SEARCH

SEARCH BY CITATION

Keywords:

  • positioning method;
  • fingerprinting database;
  • cellular networks;
  • mobile advertisement;
  • context-aware;
  • ad targeting

SUMMARY

  1. Top of page
  2. SUMMARY
  3. 1 INTRODUCTION
  4. 2 RELATED WORK
  5. 3 PRINCIPLE OF FINGERPRINTING-BASED POSITIONING METHOD
  6. 4 THE PROPOSED APPROACHING DETECTION METHOD
  7. 5 THE CONTEXT-AWARE AD TARGETING METHOD
  8. 6 EXPERIMENTAL RESULTS
  9. 7 CONCLUSION
  10. ACKNOWLEDGEMENTS
  11. REFERENCES

With the rapid development of smartphones and personal tablet computers, it brings a greatly growing rate of ubiquitous applications for location-based services (LBS). One famous LBS is the mobile advertisement. A mobile advertisement system brings benefits and opportunities among users, service providers, and advertisers. In this paper, we propose a personalized context-aware mobile advertisement system (PCA-MAS) over cellular networks, which contains two new techniques called (i) approaching detection method (ADM) and (ii) context-aware ad targeting method (CAADTM). ADM can find some point of interests that a user is approaching; CAADTM pushes advertisements that satisfy user's requirement based on the user's context, that is, user's profile, current time, current position, and so on. Our experimental results show that (i) ADM has the good hit rate to determine those point of interests that a user is approaching within the 150-m radius of the approaching range, and (ii) CAADTM has the good hit rate of finding appropriate advertisements that a user prefers through the favorite content table filtering, the annoying content table filtering, and the advertisement clicking feedback. Copyright © 2013 John Wiley & Sons, Ltd.

1 INTRODUCTION

  1. Top of page
  2. SUMMARY
  3. 1 INTRODUCTION
  4. 2 RELATED WORK
  5. 3 PRINCIPLE OF FINGERPRINTING-BASED POSITIONING METHOD
  6. 4 THE PROPOSED APPROACHING DETECTION METHOD
  7. 5 THE CONTEXT-AWARE AD TARGETING METHOD
  8. 6 EXPERIMENTAL RESULTS
  9. 7 CONCLUSION
  10. ACKNOWLEDGEMENTS
  11. REFERENCES

In recent years, there are several ubiquitous applications developed on popular smartphones based on location-based services (LBS) [1]. One famous application of LBS is mobile advertisement. The aim of the mobile advertisement is to deliver appropriate advertisement messages to the targeted users' mobile devices, which is achieved using the so-called advertisement targeting method (ad targeting method) [2]. Mobile advertisement can create benefits among users, service providers, and advertisers, for which (i) users can catch advertisement messages that they want to receive, (ii) service providers can earn money by providing the corresponding services of transmitting advertisements, and (iii) advertisers promote their activities or products through the platform of mobile advertisement. Some reports about market analysis in [3-5] is shown that (i) mobile advertising is more effective than some online advertisings, (ii) market of mobile advertising could be bigger than that of Internet advertising, and (iii) Facebook will make $1.2bn annually from mobile advertisings. From these three viewpoints, mobile advertising could become a killer application on the market at the next generation.

A mobile advertisement system can provide advertisements to a user's mobile device through the Internet. One application domain of the mobile advertisement system is having advertisements during traveling. When a user travels in some area, the mobile advertisement system provides information or advertisements of restaurants, human scenery and shops, and so on, which are near a user. These restaurants, human scenery and shops are called points of interests (POIs) in the traveling centric mobile advertisement. To provide suitable advertising to user, one of the important characteristics for mobile advertisement system is the positioning technique.

In Figure 1(a), a system can find POIs in the surrounding area of a user's current location and then, send suitable advertising to the user. A more precise way is to find POIs based on not only user's current location but also user's moving direction, and trajectory. For example, in Figure 1(b), a more precise set of POIs that are located inside the red circle, which is derive based on user's current position, moving direction, and trajectory, should be pushed to the user. To achieve the goal, this paper proposes a personalized context-aware mobile advertisement system (PCA-MAS).

image

Figure 1. Illustrated examples of adopting (a) the traditional positioning method and (b) the proposed ADM for the traveling centric mobile advertisement.

Download figure to PowerPoint

The proposed PCA-MAS contains two new techniques, which is called (i) approaching detection method (ADM) and (ii) context-aware ad targeting method (CAADTM). ADM is used to detect corresponding POIs that a user is approaching in the future moving trajectory; CAADTM tries to push some appropriate advertisements to a user based on the results of ADM. For an instance that is depicted in Figure 1(b), let Jack be traveling along the Anping District for his first time and want to have some information, that is, advertisements, of Tainan traditional snacks and human scenery in this area. When Jack is approaching some POIs, PCA-MAS can detect some POIs that he is approaching through ADM. Then PCA-MAS selects appropriate advertisements sent to Jack based on Jack's context, including current time, user's profile, and so on, through CAADTM.

To reduce spamming advertisements that are sent to a user and reach the aim of personalization in the mobile advertisement, CAADTM has four parts, which are the following: (i) favorite content table filtering, (ii) annoying content table filtering, (iii) time context reranking, and (iv) advertisement clicking feedback. To effectively find advertisements that can satisfy a user's interest, the technique of favorite content table filtering gives scores to those types of POIs that a user prefers. To avoid the problem of recommending advertisements of a POI to the user on which time, this POI is not opening, time context reranking justifies whether current time is in the opening period of POIs that a user is approaching or not. The advertisement clicking feedback computes the weight for each type of POI using the rate of clicking advertisements. In other words, a user may be interested in some types of POIs when the user clicks the advertisements of them. Thus, our proposed PCA-MAS gives some weight to raise the scores of those types of POIs that the user has clicked. Finally, PCA-MAS finds the appropriate advertisements to a user according to the scores of POIs.

The rest of the paper is organized as follows. Section 2 introduces related work about the fingerprinting-based positioning methods and mobile advertisement systems. Section 3 illustrates principles of the fingerprinting-based positioning method. Section 4 introduces the ADM. Section 5 presents the CAADTM of PCA-MAS. Section 6 shows the PCA-MAS system and the experimental environment and analyzes experimental results. Finally, the conclusion remarks are given in Section 7.

2 RELATED WORK

  1. Top of page
  2. SUMMARY
  3. 1 INTRODUCTION
  4. 2 RELATED WORK
  5. 3 PRINCIPLE OF FINGERPRINTING-BASED POSITIONING METHOD
  6. 4 THE PROPOSED APPROACHING DETECTION METHOD
  7. 5 THE CONTEXT-AWARE AD TARGETING METHOD
  8. 6 EXPERIMENTAL RESULTS
  9. 7 CONCLUSION
  10. ACKNOWLEDGEMENTS
  11. REFERENCES

Two research domains about mobile advertising are (1) positioning technique and (2) advertisement filtering technique. Several positioning methods have been proposed based on the cellular networks, including angle of arrival [6], time of arrival [7], time difference of arrival [8], enhanced observed time difference [9], timing advance [10], and so on. However, the aforementioned methods not only need the help of cellular networks operators but also need extra hardware installed in the base station. Moreover, research results depicted in [11] found that the positioning accuracy of the fingerprinting-based positioning method is higher than that of the aforementioned methods. The concept of the fingerprinting-based positioning technique is to collect some data such as cellular information and then to use these data to build a fingerprint database [12, 13]. When a user would like to have positioning, user's cellular information is sent to the server side to find matched fingerprint records between the user's cellular information and the fingerprinting database. Finally, the matched record contains location information indicating the users' location.

In the traveling centric mobile advertisement, there are two main types of transmitting advertisement messages to the targeted user's mobile device: (i) pull-based ad targeting method [14] and (ii) push-based ad targeting method [15-17]. The pull-based ad targeting method is used to transmit corresponding advertisements to a user when the user sends the searching constraints of advertisements. The searching constraints could be (i) location and (ii) types of POIs, such as restaurants, human scenery, and so on. Location can be user's current location or the searched area. For example, a user searches the advertisements of restaurants that are in the surrounding area of the Anping Fort, Tainan City. In [14], authors proposed the mobile advertisement system based on the pull ad targeting method over Bluetooth networks. The pull ad targeting method only provides corresponding advertisements to a user when a user wants to obtain information of POIs.

However, the pull-based ad targeting method cannot actively transmit some types of advertisements to a user, in which advertisements may attract user's interest. The push-based ad targeting method actively provides advertisements to the user according to user's current location. For example, when a user enters into the surrounding area of the Anping Fort, Tainan City, the mobile advertisement system sends the corresponding advertisements to the user, in which these advertisements contain information about those restaurants in that area. But, it may cause spamming advertisements to a user when the push-based ad targeting method only takes user's current location as the criteria. In [16], authors proposed the mobile advertisement system using the technique of Bluetooth positioning and Wireless Application push. The main idea of the study is to calculate the user's location via the Bluetooth sensors and send the advertisements of some POIs, which are nearby, to the user. However, the aforementioned push-based ad targeting method is only based on the user's location. It will cause the problem of sending spamming advertisements to the user.

To reduce the spamming advertisements and to achieve the goal of sending personalized advertisements to a user, some researches proposed push methods by utilizing the user's contextual information [17-24]. In [17], authors proposed a data push scheduling scheme using the user's preference model. Authors of [17] analyzed user's preference on each time of a day, that is, the user's historic behavior is computed as a preference model. However, it needs to observe each user for a long period and collects enough information about their behaviors. When many users use the corresponding advertisement system, it seems an unsuitable way to construct and implement the preference model from collecting information of each user. In [18-24], authors proposed the data push scheme in the mobile advertisement system by utilizing the user's contextual information, including the user's profile, location, and historic behavior.

The aforementioned mobile advertisement systems did not consider the user's annoying content, the opening period of POIs, and the popular rate of each POI. It will cause some problems: (i) Sending inappropriate advertisements to a user could result in a situation that the user may not like these types of pushed advertisements. (ii) Sending the advertisements of POIs to a user at unsuitable time. (iii) Missing potential advertisements that may attract a user: when the mobile advertisement system only considers the user's favorite contents, it cannot push those popular POIs to the user, in which these advertisements are not in the user's favorite content table and are not in the user's annoying table either. In other words, it cannot create any opportunity for attracting a user's interest for these POIs, which are neither in the user's favorite content table nor in the user's annoying content table, but are popular in other users. To overcome these problems, CAADTM was proposed in this paper, and detailed description is given in Section 5.

3 PRINCIPLE OF FINGERPRINTING-BASED POSITIONING METHOD

  1. Top of page
  2. SUMMARY
  3. 1 INTRODUCTION
  4. 2 RELATED WORK
  5. 3 PRINCIPLE OF FINGERPRINTING-BASED POSITIONING METHOD
  6. 4 THE PROPOSED APPROACHING DETECTION METHOD
  7. 5 THE CONTEXT-AWARE AD TARGETING METHOD
  8. 6 EXPERIMENTAL RESULTS
  9. 7 CONCLUSION
  10. ACKNOWLEDGEMENTS
  11. REFERENCES

In this Section, we depict the principle of the fingerprinting-based positioning method. The main idea of our proposed positioning technique is to collect cellular information of POIs and establish the fingerprinting database on a server initially; next, user's related cellular information is sent to the server side when the user would like to find some POIs that the user is approaching; then, the fingerprint server finds matched fingerprint records between user's cellular information and the fingerprinting database. The user's related cellular information contains neighboring base station's (BS) Cell Identification (ID), Sector Identification (Sector ID), and received signal strength indication (RSSI). The key point of our idea is to find some approaching POIs based on the user's moving direction and trajectory, but not the user's position, that is, user's current longitude and latitude. Three parts of the fingerprinting-based positioning technique are as follows: (1) cellular-related information collection, (2) fingerprinting database establishment, and (3) fingerprinting-based positioning method.

3.1 Cell-ID's related information collection

The first step of the fingerprinting-based positioning technique is to collect Cell-ID's related information of neighboring BSs in each POI, that is, human scenery, food vendors, and so on, to establish a fingerprinting database at the server side. Some tools such as smartphones help us to collect these Cell-ID's related information, which contains Cell-ID and RSSI. To enhance the positioning accuracy in our proposed ADM, it needs to measure enough Cell-ID's related information in each POI for a period of time.

3.2 Fingerprinting database establishment

In this part, we introduce how PCA-MAS establishes the fingerprinting database using collected relevant information.

In the phase of collecting Cell-ID's related information, up to six Cell-IDs and corresponding RSSIs of neighboring BSs can be detected in a mobile device. The number ‘six’ is the upper bound offered by the android API over cellular networks. We develop a cellular information collecting tool based on the SDK of the android platform to obtain these six Cell-IDs' related information. Six neighboring BSs' Cell-IDs are selected from the top six strongest RSSI among neighboring BSs. Additionally, each POI's location is measured using the Global Positioning System (GPS). Thus, each Cell-ID's related information contains the GPS location, neighboring BS's Cell-ID, Sector ID, and RSSI. Each fingerprint record includes the Cell-ID set and the corresponding RSSI. After the Cell-ID's related information is collected, it needs to calculate the derived attributes of each fingerprint record of the transformed fingerprinting database, that is, mean of RSSI and standard deviation of RSSI, which is depicted in Figure 2. The derived information can be used for helping filter unnecessary fingerprint records and obtain higher accuracy of determining whether a user is approaching a POI or not.

image

Figure 2. The format of the transformed fingerprinting database.

Download figure to PowerPoint

3.3 Fingerprinting-based positioning method

The main idea of the fingerprinting-based positioning method is to use a pattern matching algorithm for finding the best matching record in the fingerprinting database. Then, it looks up the corresponding GPS location as the user's location from the best matching record. In Figure 3, the fingerprinting information and fingerprinting records contain BSs' information, including Cell-ID, Sector ID, and corresponding RSSI.

image

Figure 3. The execution configuration of the fingerprinting-based positioning method.

Download figure to PowerPoint

4 THE PROPOSED APPROACHING DETECTION METHOD

  1. Top of page
  2. SUMMARY
  3. 1 INTRODUCTION
  4. 2 RELATED WORK
  5. 3 PRINCIPLE OF FINGERPRINTING-BASED POSITIONING METHOD
  6. 4 THE PROPOSED APPROACHING DETECTION METHOD
  7. 5 THE CONTEXT-AWARE AD TARGETING METHOD
  8. 6 EXPERIMENTAL RESULTS
  9. 7 CONCLUSION
  10. ACKNOWLEDGEMENTS
  11. REFERENCES

Approaching detection method adopts the manner to collect Cell-ID's related information and calculate the derived parameters, that is, mean of RSSI and standard deviation of RSSI, to help decide whether a user is approaching some POIs or not. Because some proposed fingerprinting-based positioning methods need to collect the cellular information about the whole experimental field, the cost is too high to implement fingerprinting-based positioning methods for the LBS in the whole area, that is, the whole area of a country. We propose ADM to solve the aforementioned problem for some specific purposes, that is, mobile advertisement system and tourist guidance system. In contrast with traditional fingerprinting-based positioning methods over cellular networks, ADM does not position a user's location, but provides a way to compute the existed probabilities of POIs that a user may approach on the basis of the established Cell-ID's related information at server's database.

Approaching detection method adopts the basic concept of the pattern matching method, in which some symbols used in ADM are defined in Table 1, and three components that ADM has are as follows: (1) neighboring Cell-ID mapping, (2) RSSI-scale mapping, and (3) POIs approaching decision.

Table 1. Symbols definition for approaching detection method.
SymbolsDefinition
  1. POI, point of interest; RSSI, received signal strength indication.

Ncoincidence_cellIDNumber of matched records after comparing the Cell_ID between the user side and the database side.
Nreceived_user_cellIDNumber of received Cell-ID's information at the user side.
Papproaching_cellIDThe probability of approaching the POI.
Papproaching_cellIDRSSIThe probability of POIs that a user may approach by comparing the similarity of RSSI among each matching Cell-IDs between a user's cellular information and fingerprint records.
PapproachingThe probability that a user is approaching to POI based on Cell-ID mapping and RSSI-scale mapping.
NRSSI_scale_mappingNumber of cell's RSSI belongs to the range of the fingerprint record.

4.1 An illustration of approaching detection method

In Figure 4, there are totally N POIs in the fingerprinting database, and there are at most six Cell-IDs and its corresponding RSSI detected by the cellular phone in a POI per second. We measure Cell-ID's related information in each POI for a period of time, in which the default time length is 30 s. For example, we collect Cell-ID's related information in POI1 from T1 to T30.

image

Figure 4. Illustrated fingerprinting database transformation of approaching detection method. RSSI, received signal strength indication; POI, point of interest; SD, standard deviation.

Download figure to PowerPoint

After measuring the Cell-ID's related information, it needs to compute the derived information, that is, mean of RSSI and standard deviation of RSSI. Both of them are defined as follows:

  1. Mean of RSSI: Given one neighboring Cell-ID, let there be different RSSI. The value i is set from 1 to m that are detected by the cellular phone in the specific location for the time length T. Mean of RSSI (MeanRSSI(Cell-ID)) is the average of those RSSIi of one neighboring Cell-ID, which is given in Equation (1).
    • display math(1)
  2. Standard deviation (SD) of RSSI: Let there be different RSSIi. The value i is set from 1 to m, which is one neighboring Cell-ID and the derived mean of RSSI (MeanRSSI(Cell-ID)) in the specific location for the time length T. Standard deviation of RSSI (SDRSSI(Cell-ID)) of RSSI of one neighboring Cell-ID is given in Equation (2).
  • display math(2)

Figure 4 depicts the table of the transformed fingerprinting database after calculating the derived attributes. For example, there are four different neighboring Cell-IDs, including 51331, 51957, 51958, and 51992 in the same location1 of POI1; their corresponding attributes are also depicted accordingly.

Three parts of ADM are (1) neighboring Cell-ID mapping, (2) RSSI-scale mapping, and (3) POIs approaching decision. In Figure 5, there are five Cell-ID sets received by the cellular phone of a user, that is, 51331, 51957, 51958, 51992, and 51994. Each Cell-ID set contains Cell-ID and RSSI. After transmitting Cell-ID's related information detected by the cellular phone of a user to the fingerprinting server, ADM does the process of neighboring cell-ID mapping first. Neighboring cell-ID mapping finds the number of matches of cell-ID set between a user and the fingerprinting database, and the number of matches is denoted as Ncoincidence_cellID. In Figure 5, there are four Cell-IDs received by the cellular phone of a user that are identical to that of the fingerprint record, hence Ncoincidence_cellID equals 4. Then, ADM computes the probability of approaching a POI (Papproaching_cellID) according to the parameter Ncoincidence_cellID. The denominator of Papproaching_cellID represents the number of Cell-IDs received by the cellular phone of a user, and the numerator represents the number of matches of the Cell-ID set between a user and the fingerprinting database. In Figure 5, Papproaching_cellID equals 0.8.

image

Figure 5. An illustration of using the approaching detection method. RSSI, received signal strength indication; POI, point of interest; SD, standard deviation.

Download figure to PowerPoint

After processing neighboring cell-ID mapping, ADM executes the process of RSSI-scale mapping. In Figure 5, RSSI-scale mapping mainly compares whether the user's RSSI belongs to the RSSI range of the fingerprinting database or not. The RSSI range of each Cell-ID is within (i) mean of RSSI (Mean of RSSI) minus SD of RSSI (SD of RSSI) and (ii) mean of RSSI (Mean of RSSI) plus SD of RSSI (SD of RSSI). Then ADM calculates Papproaching_cellIDRSSI, which represents the probability of POIs that a user may approach by comparing the similarity of RSSI among each matching Cell-IDs between a user's cellular information and fingerprint records of the fingerprinting database. In Figure 5, Papproaching_cellIDRSSI equals 0.5.

Finally, ADM operates the POIs approaching decision after having neighboring Cell-ID mapping and RSSI-scale mapping. In Figure 5, Papproaching is the product of Papproaching_cellID and Papproaching_cellIDRSSI, in which Papproaching represents the probability in which a user is approaching a POI based on Cell-ID mapping and RSSI-scale mapping. ADM sets the threshold of the probability of approaching a POI, if Papproaching is equal to the threshold or greater than the threshold, ADM puts the corresponding POI with Papproaching into the candidate list, that is, ADM will observe the probability of a user's approaching the POI for a period of time T. From our observation, the value of T should be set between 15 s to 45 s. If the value T is too small, ADM is hardly to find the correct POIs that a user is approaching. The reason is that the sampled signals of the user device are not stable enough during the sampling duration T, and thus many POIs that satisfy the searching criteria based on the unstable signals are imprecisely selected. If the value T is too big, the ADM may be inefficient because the user has to wait for longer time to obtain the correct POIs that the user is approaching. In addition, ADM can be used on different status of the user, that is, pedestrian, biker, or driver, as long as the fingerprinting database is made accordingly. That is, when a pedestrian/biker/driver-based fingerprinting database is made, ADM then can be used for pedestrians/bikers/drivers. During the time interval T, if there is the increasing or consistent trend of the probability that a user may approach the POI, ADM determines that a user is approaching this POI, which is depicted in Figure 5.

4.2 Formalization of approaching detection method

4.2.1 The execution of approaching detection method

Three steps of ADM are as follows: (1) neighboring Cell-ID mapping, (2) RSSI-scale mapping, and (3) approaching POI decision, which are depicted in Figure 6. In the first step, ADM maps the neighboring Cell-IDs received by the cellular phone of a user into that of the fingerprinting database. Then, it computes the probability that a user may approach the POI according to the matches of Cell-ID between a user and each fingerprint record of the fingerprinting database. In the second step, ADM computes the probability that a user may approach the POI using the RSSI-scale mapping among matched Cell-IDs. In the third step, ADM finds the candidate POIs that a user may approach using the probability of Cell-ID mapping and the probability of RSSI mapping. Then, ADM observes candidate POIs for a period of time. If there is the growing or consistent trend of the probability within the observing time interval, ADMoP determines the POI that a user is approaching.

image

Figure 6. The execution flow chart of the approaching detection method. POI, point of interest; RSSI, received signal strength indication.

Download figure to PowerPoint

4.2.2 Neighboring Cell-ID mapping

In this part, ADM computes the probability that a user may approach a POI, which can avoid the problem due to the reason of changing BSs, that is, a new BS is installed or a BS goes wrong. First, ADM finds the number of matches of neighboring Cell-IDs between a user and each fingerprint record of the fingerprinting database and compute the probability that a user approaches a POI (Papproaching_cellID), which is given in Equation (3). In Equation (3), Nreceived_user_cellID means the number of neighboring Cell-IDs that is received by the cellular phone of a user; Ncoincidence_cellID represents the number that Cell-IDs of a user's cellular information are identical to fingerprint records of the fingerprinting database.

  • display math(3)

4.2.3 Received signal strength indication scale mapping

After processing neighboring Cell-ID mapping, ADM computes the probability that a user may approach a POI through the corresponding RSSI scale mapping. By taking the corresponding RSSI of each neighboring Cell-ID into consideration, it can help filter inappropriate POIs that a user does not approach. Equation (4) depicts calculation of the probability that a user approaches the POI (Papproaching_cellIDRSSI). Papproaching_cellIDRSSI computes the probability of RSSI similarity between the user's cellular information and fingerprint records of the fingerprinting database.

  • display math(4)

4.2.4 Point of interests approaching decision

After processing neighboring Cell-ID mapping and RSSI-scale mapping, ADM computes the probability that a user may approach a POI (Papproaching), in which Papproaching is the product of Papproaching_cellID and Papproaching_cellIDRSSI. In Equation (5), if Papproaching is equal to or greater than the threshold, ADM selects the corresponding POIs as the candidate, to which a user probably approaches. To accurately assure that a user approaches the candidate POIs, ADM observes each candidate POI for a period of time T. If there exists consistent or increasing trend of Papproaching for the time interval T, ADM determines that a user is approaching the POI. The judgment rule is given in Equation (6).

  • display math(5)
  • display math(6)

5 THE CONTEXT-AWARE AD TARGETING METHOD

  1. Top of page
  2. SUMMARY
  3. 1 INTRODUCTION
  4. 2 RELATED WORK
  5. 3 PRINCIPLE OF FINGERPRINTING-BASED POSITIONING METHOD
  6. 4 THE PROPOSED APPROACHING DETECTION METHOD
  7. 5 THE CONTEXT-AWARE AD TARGETING METHOD
  8. 6 EXPERIMENTAL RESULTS
  9. 7 CONCLUSION
  10. ACKNOWLEDGEMENTS
  11. REFERENCES

The proposed CAADTM adopts the data push mechanism based on the user's context, that is, user's preference of each type of advertisements, user's annoying content of advertisements and current time, user's current location, and so on. Six issues that exist in CAADTM are the following: (i) configuration of mobile advertisement database and user's profile, (ii) favorite content table filtering, (iii) annoying content table filtering, (iv) time context reranking, (v) advertisement clicking feedback, and (vi) the execution flow chart of CAADTM. Table 2 depicts some symbols' definitions that are used in the CAADTM.

Table 2. Symbols definition for context-aware ad targeting method.
SymbolsDefinition
CTCurrent time
OTPRange of the opening time
iPOIThe ith POI
CL_POICandidate list of POIs for advertisement
ACRAdvertisement clicking rate
NoC_ iPOINumber of clicking of ith POI
NoC_TotalTotal number of clicking

5.1 Configuration of mobile advertisement database and user's profile

Let the two main types of POIs exist in the illustrated example of using our proposed CAADTM for the mobile advertisement system be (i) human scenery and (ii) traditional snacks in Tainan City, which is depicted in Figure 7. Figure 7(a) depicts that there are six types of human scenery in Tainan City; Figure 7(b) describes that there are totally 16 types of traditional snacks in Tainan City. Let there be totally nine records in the mobile advertisement database. Each record depicts attributes of a POI. These attributes are as follows: (i) name of POI, (ii) location, (iii) type, (iv) description of POI, and (v) popular rate. The popular rate of a POI represents the preferred rate of the POI for all users. For example, the popular rate ‘0.05’ of POI5 represents that 5% of all users prefer this POI, and these users had used the proposed application in mobile advertisement.

image

Figure 7. Illustrated two main types of point of interests for explaining the proposed context-aware ad targeting method: (a) human scenery in Tainan City and (b) traditional snacks in Tainan City.

Download figure to PowerPoint

Before transmitting the request for the advertisements of POIs, a user needs to set the corresponding profile, which contains the type of favorite contents and annoying contents. In Figure 8(a), there are five types of traditional snacks in Tainan City that a user prefers, which is depicted in the favorite content table. According to the rank of the favorite content table of a user, CAADTM gives the corresponding score to each type of traditional snacks. For example, if ‘Fried-Spanish Mackerel Thick Soup’ is the most favorite type of traditional snacks for a user. Then CAADTM gives 5 as a score to the most favorite type of traditional snacks of a user, that is, ‘Fried-Spanish Mackerel Thick Soup’. Similarly, CAADTM gives the corresponding negative scores to each annoying type of traditional snacks using the annoying content table depicted in Figure 8(b). For example, ‘Shrimp Roll’ is the most annoying type of traditional snacks of a user. CAADTM gives the highest negative score to the most annoying type, that is, ‘Shrimp Roll’.

image

Figure 8. An illustration of a user's profile: (a) favorite content table and (b) annoying content table.

Download figure to PowerPoint

5.2 Favorite content table filtering

To select appropriate advertisement of some POIs that a user is approaching, we take the user's preferred contents into consideration. The favorite content table filtering mainly gives the corresponding score to each type of some POIs that a user is approaching based on the favorite content table of a user. Initially, the score of each POI is shown as Equation (7), and the popular rate represents the preferred rate of a POI among all users. The score of each type of POI is represented as the degree of the preferred rate. In other words, POI is preferred much by a user if the corresponding score becomes higher. In Figure 9, the number of preferred contents of a user is denoted as N(fv). The favorite content table filtering maps the favorite type of traditional snacks depicted in the favorite content table into the same type of some POIs at first, to which POIs, a user is approaching. Then, CAADTM gives the corresponding scores to those POIs. For example, ‘Fried-Spanish Mackerel Thick Soup’ is the most favorite type of traditional snacks of a user. Then, the favorite content table filtering gives score, that is, N(fv), to both ‘Cheng Chi Fired-Spanish Mackerel Thick Soup’ and ‘Chen Chi Fried-Spanish Mackerel Thick Soup’, in which both of them belong to the type ‘Fried-Spanish Mackerel Thick Soup’.

image

Figure 9. Formalization of the favorite content table filtering.

Download figure to PowerPoint

  • display math(7)

5.3 Annoying content table filtering

After processing the favorite content table filtering, CAADTM operates the annoying content table filtering. To filter inappropriate advertisements that a user does not like, the annoying content table filtering mainly gives the corresponding negative score to some POIs based on the annoying content table of a user. In Figure 10, the number of annoying contents of a user is denoted as N(ANN). The annoying content table filtering gives the corresponding negative score to the same type of POIs after mapping the favorite content table and those POIs that a user may approach. The negative score of each type of POIs represents the degree that a user does not like. For example, a user does not like ‘Milkfish Ball/Milkfish Congee’, the annoying content table gives the negative score, that is, −1, to ‘Yung Chi Milkfish Ball’.

image

Figure 10. Formalization of the annoying content table filtering. POI, point of interest.

Download figure to PowerPoint

5.4 Time context reranking

After processing the favorite content table filtering and the annoying content table filtering, CAADTM operates the time context reranking. To avoid recommending advertisements of a POI to the user on which time this POI is not opening, the time context reranking mainly justifies whether current time is in the range of the opening period of a POI or not, which is given in Equation (8). In Figure 11, assume current time is not in the range of the opening period of ‘A Tieh Fried Eel Noodles’, CAADTM deletes ‘A Tieh Fried Eel Noodles’ from the candidate list.

image

Figure 11. An illustration of time context reranking. POI, point of interest.

Download figure to PowerPoint

  • display math(8)

5.5 Advertisement clicking feedback

To accurately transmit appropriate advertisements to a user, the advertisement clicking feedback is the important component to decide which types of advertisements a user may prefer. The advertisement clicking feedback mainly computes the number of clicking advertisements of POIs and adjusts the score of each POI based on the number of clicking advertisement. The advertisement clicking feedback computes the weight of this type of traditional snacks and gives the new score to the same type of POIs, which a user may approach. Equation (9) depicts the computation of advertisement clicking rate of each type of POIs. A higher clicking rate of a POI represents that a user prefers this POI much. Then, the advertisement clicking feedback computes the weight of each type of POIs according to the number of clicking the same type of advertisement, which is given in Equation (10). In our design, the weight of each type of POI is given 1. If a user never clicks this type of POI, the weight of the POI is not changed. However, if the advertisement of the type of a POI is clicked by a user, the score of the same type of POIs will be multiplied by the weight, which is given in Equation (11). If advertisements of some types of POIs are clicked by a user several times, then it represents that a user is interested in these types of POIs. Therefore, CAADTM raises the score of the corresponding types of POIs. Then, a user will see these types of POIs in the top of advertisements.

  • display math(9)
  • display math(10)
  • display math(11)

In Figure 12, a user clicks the advertisements of ’Cheng Chi Fried-Spanish Mackerel Thick Soup’ five times, it represents that a user is interested in the type ‘Fried-Spanish Mackerel Thick Soup’. Weight (Fried-Spanish Mackerel Thick Soup) equals (1 + 5/8 ), which represents the clicking rate of the advertisements of ‘Fried-Spanish Mackerel Thick Soup’. Score (Fried-Spanish Thick Soup) represents a new score of the type ‘Fried-Spanish Mackerel Thick Soup’ based on the number of clicking advertisement.

image

Figure 12. An illustration of the advertisement clicking feedback.

Download figure to PowerPoint

5.6 The execution flow chart of context-aware ad targeting method

Figure 13 gives the execution flow chart of the CAADTM. The input of CAADTM is some POIs that a user is approaching, and the input is produced using the ADM. Let (i) N be the number of POIs that a user is approaching and (ii) the POI set store POIs that a user is approaching. CAADTM gives a corresponding score to this POI through (i) favorite content table filtering, (ii) annoying content table filtering, (iii) time context reranking, and (iv) advertisement clicking feedback, respectively. The favorite content table gives the score to some POIs. These types of POIs belong to the favorite content table of a user. Then, CAADTM operates the annoying content table filtering. The annoying content table filtering gives the negative score to some types of POIs, which a user does not like. After processing the annoying content table filtering, CAADTM computes scores of each POI using Equations (8) and (9), and ranks POIs according to the scores of these POIs, which a user is approaching. The top of the ranking POIs represents that it is the most favorite type of POI for a user. Then, CAADTM operates the time context reranking, which deletes those POIs from the list if current time is not in the range of the opening periods of these POIs. The advertisement clicking feedback computes the new score of a POI based on the number of clicking advertisement. When the POI set is empty, it represents that it is the end of the loop of giving the corresponding score to the advertisement of each POI. Finally, CAADTM deletes those POIs that have negative scores from the list. Advertisements of the POIs in the list will be sent to the user.

image

Figure 13. The execution flow chart of context-aware ad targeting method.

Download figure to PowerPoint

6 EXPERIMENTAL RESULTS

  1. Top of page
  2. SUMMARY
  3. 1 INTRODUCTION
  4. 2 RELATED WORK
  5. 3 PRINCIPLE OF FINGERPRINTING-BASED POSITIONING METHOD
  6. 4 THE PROPOSED APPROACHING DETECTION METHOD
  7. 5 THE CONTEXT-AWARE AD TARGETING METHOD
  8. 6 EXPERIMENTAL RESULTS
  9. 7 CONCLUSION
  10. ACKNOWLEDGEMENTS
  11. REFERENCES

6.1 System architecture

The PCA-MAS was implemented for experiment, and the system architecture of PCA-MAs is depicted in Figure 14. Three main parts of the PCA-MAS are as follows: (i) an application in the client, (ii) the mobile advertisement system, and (iii) the fingerprinting-based positioning technique. The application on the client includes three parts: (i) context information setting, (ii) neighboring Cell-ID's related information detection, and (iii) the number of clicking each type of advertisement records. The fingerprinting-based positioning technique contains (i) information collection for the fingerprinting database and (ii) ADM. The mobile advertisement system contains the CAADTM.

image

Figure 14. The system architecture of the proposed personalized context-aware mobile advertisement system. RSSI, received signal strength indication; POI, point of interest.

Download figure to PowerPoint

In the application of the client, context information setting provides the interface of setting the user's profile, that is, favorite contents and annoying contents, and getting current time. The neighboring Cell-ID's related information detection can obtain the Cell-ID and the corresponding RSSI of the neighboring BSs. The number of clicking each type of advertisement computes the number of clicking different types of advertisements.

In the fingerprinting-based positioning technique, the information collection mainly handles measuring the Cell-ID's related information (e.g., Cell-ID and RSSI). The novel ADM finds some POIs that a user may approach based on neighboring BSs' information.

In the mobile advertisement system, the CAADTM pushes the appropriate advertisements to a user based on the user's profile, that is, the favorite content table, the annoying content table, and the current time. When a user uses the proposed application in the mobile phone, it sends the context information of a user to the mobile advertisement server, that is, a user's profile, current time, and the Cell-ID's related information. Then, the mobile advertisement server selects appropriate advertisements of POIs through ADM and CAADTM.

6.2 Experiment results

In this section, we discuss two phases of the PCA-MAS: (i) detecting the POIs that a user is approaching using the ADM and (ii) filtering and then pushing suitable advertisements to the user using the CAADTM. We present the experimental environment and analyze the experimental results of using our proposed ADM and CAADTM. The experimental environment is introduced in the first part, and the second part gives analysis of experimental results.

6.2.1 Experiment 1: approaching detection method

In this part, the experimental results of using the ADM in the urban environment show that our proposed ADM can effectively find POIs that a user is approaching. In this experiment, there are totally 22 POIs in the experiment field, and range of the field is about 5 × 2 km2. Twenty-two POIs were selected in the experimental area because all of them are historic spots or other interesting points. Time to collect related cellular information at each POI is about 30 s to obtain the information quickly. The mobile device we used for information collecting is HTC Legend smartphone (High Tech Computer Corporation,Taiwan, R.O.C.), and the operation system is Android 2.1. A server (ACER, Taiwan, R.O.C.) is setup on a personal computer with the Intel Core 2 Quad Q8300 CPU and 4 G RAM. A user was approaching a POI on foot when the experiment was operated.

To explain the hit rate of finding POIs that a user is approaching in the experimental results clearly, we define some terminologies as follows:

  1. Targeted range: In Figure 15, when a user is moving, the moving vector of a user is shown as the red arrow line, which represents the moving direction and the moving speed of a user. We can infer that a user will be in the specific location in 60 s by computing ‘the moving speed of a user ×60 (seconds)’. This location of a user must be on the border of the targeted range, which is represented as the blue circle. We infer that the center of the targeted range is at a distance, for example, (the moving speed of a user ×60 (seconds) + radius of the targeted range), of the user's current location. The targeted range denotes the range of the area that a user will pass through in the specific period.
  2. Tightly targeted POIs: In Figure 15, when a user is approaching the targeted range on his moving vector during the specific period, ADM finds POIs that a user is approaching. These POIs in the targeted range are called the tightly targeted POIs that a user is approaching. For example, POI1 and POI2 depicted in Figure 15 are the tightly targeted POIs.
  3. Loosely targeted POIs: In Figure 15, ADM may find some POIs that are not in the targeted range. These POIs are adjacent to the targeted range. These POIs are called loosely targeted POIs. For instance, POI3, POI4, POI5, and POI6 depicted in Figure 15 are loosely targeted POIs.
image

Figure 15. An illustrated example of defining terminologies for experiments using approaching detection method.

Download figure to PowerPoint

We discuss (i) relationship between probability threshold and the hit rate of finding POIs that a user is approaching, and (ii) comparison between our proposed ADM and the detection method only using the neighboring Cell-ID mapping (DM_CellMapping).

6.2.1.1 Experiement 1.1: experimental environment

In Figure 16(a), the area is selected as the experimental field, which is the Central and West District of Tainan City. We choose some POIs that are in the Central and West District of Tainan City. In each POI, we collect Cell-ID's related information for a time length, and the default time length is set to 60 s using the collecting cellular information method depicted in Figure 16(b). Time length is set to 60s because spending more time to collect related information can enhance the calculation results accuracy. Thereafter, we calculate related attributes from collected cellular information of each POI, including mean received signal strength (RSS) and SD of RSS for ADM.

image

Figure 16. The experimental field of approaching detection method in the urban environment.

Download figure to PowerPoint

6.2.1.2 Experiment 1.2: experimental results

6.2.1.3 Part1: relationship between probability threshold and the hit rate of finding point of interests that a user is approaching

In this part, we show relationship between probability threshold and the hit rate of finding POIs that a user is approaching. In Figure 17, it is obvious to observe that the average hit rate of finding POIs that a user is approaching is better when the probability threshold equals 0.2 in different radius of the targeted range. When the radius of the targeted range is 50m, it has lower possibility of determining all tightly targeted POIs that a user is approaching. When the radius of the targeted range increases, the possibility of determining all tightly targeted POIs that a user is approaching enhances. If the radius of the targeted range equals 150 m, our proposed ADM has the great hit rate with the probability threshold ‘0.2’. It implies that our proposed ADM has the great hit rate with the 150-m radius targeted range in this experiment. The probability threshold implies how many POIs can be found in the targeted range. In Figure 17, the hit rate is decreased when the probability threshold is increased because a larger probability threshold results in more POIs that can be found. These POIs may not be the correct ones because the ADM method calculates some related cellular information, for example, RSSI, in which the cellular information is not exactly accurate. For this reason, even though more POIs can be found when the probability threshold is set to be higher, the hit rate of finding POIs that a user is approaching may be decreased, especially when the targeted range becomes larger.

image

Figure 17. Relationship between probability threshold and the average hit rate of finding point of interests that a user is approaching.

Download figure to PowerPoint

6.2.1.4 Part2: comparison between our proposed approaching detection method and the detection method only using the neighboring Cell-ID mapping (DM_CellMapping)

This experiment compares ADM and the detection method only using the neighboring Cell-ID mapping (DM_CellMapping). In Table 3, the average number of loosely targeted POIs is 3 when the radius of the targeted range equals 50 meters. The average distance between these loosely targeted POIs and the center of the targeted range is about 187.3577 m in ADM. In the same condition, the average distance between loosely targeted POIs and the center of the targeted range is about 287.7277 m in DM_CellMapping, which is depicted in Table 4. In Figure 18, the average distance of loosely targeted POIs in DM_CellMapping is higher than that of ADM. It implies that DM_CellMapping only finds POIs in the large-scale range. It cannot find those loosely targeted POIs which are adjacent to the targeted range accurately. The reason is that DM_CellMapping only maps the neighboring Cell-ID between the user's cellular information and fingerprint records of the fingerprinting database. Our proposed ADM can have good hit rates of finding tightly targeted POIs, and there is no loosely targeted POI in the 150-m radius targeted range.

Table 3. Average hit rate, number of loosely targeted point of interests (POIs), average distance of approaching detection method.
Targeted rangeAverage hit rate (%)Average number of loosely targeted POIsAverage distance of loosely targeted POIs (m)
50-m radius66.673187.3577
100-m radius801136.5412
150-m radius10000
Table 4. Average hit rate, number of loosely-targeted point of interests (POIs), average distance of DM_CellMapping.
Targeted rangeAverage hit rate (%)Average number of loosely targeted POIsAverage distance of loosely targeted POIs (m)
50-m radius1008.5287.7277
100-m radius1007.5264.0355
150-m radius1007216.7704
image

Figure 18. Comparison between approaching detection method (ADM) and DM_CellMapping.

Download figure to PowerPoint

The proposed ADM is used to detect the POIs that a user is approaching. Because ADM only records POIs' location information, the cost of the fingerprint database establishment is significantly lower than that of other fingerprint-based positioning methods because cellular information collection for each point in a specific area to establish the fingerprinting database needs more time and resource. Other methods need to collect more cellular information, which may result in higher fingerprinting database establishment overhead/cost, to improve the positioning accuracy; our ADM method does not need to do it because the ADM only needs to detect which POIs a user is approaching. Furthermore, although we collect Cell's information at each POI for a period, for example, 30 s, the number of records in the server is much less than that of other fingerprint-based positioning methods.

Even if the filtering technique is used in the proposed ADM, the processing time, including searching time in the database, of the proposed ADM is very small. In addition, the purpose of the ADM method and other positioning methods is different. The purpose of ADM tries to find some POIs that a user is approaching, and the other positioning methods try to position user's location. That is, ADM does not intend to find the exact position, that is, longitude and latitude, of the user. Hence, it is inappropriate to compare the proposed ADM with other traditional cellular positioning methods, for example, angle of arrival [6], time of arrival [7], and time difference of arrival [8].

6.2.2 Experiment 2: context-aware ad targeting method

In this part, the CAADTM mainly finds appropriate advertisements of POIs that a user is approaching. After finding POIs that a user is approaching using ADM, CAADTM computes the score of each POI using four parts: (i) favorite content table filtering, (ii) annoying content table filtering, (iii) time context reranking, and (iv) advertisement clicking feedback. In the experiments, we focus on comparison between our proposed CAADTM and traditional ad targeting method. The traditional ad targeting just pushes the advertisements of POIs that a user is approaching after processing ADM.

6.2.2.1 Experiment 2.1: experimental environment

We collect information about POIs, including the POI's GPS location, name of POI, introduction about POI, and so on. The experimental environment and scenario are the same as that of ADM, that is, Central and West District of Tainan City.

6.2.2.2 Experiment 2.2: experimental results

6.2.4 Part1: comparison of favorite advertisement's hit rate between our proposed context-aware ad targeting method and traditional ad targeting method

In this part, we show that comparison of favorite advertisement's hit rate between our proposed CAADTM and traditional ad targeting method. The definition of hit rate of favorite advertisements is the number of mapping between those types of the top five selected advertisements and a user's favorite content table. We did this experiment five times. In Figure 19, the hit rate of favorite advertisements of CAADTM is always equal to or greater than that of traditional ad targeting method (Ad Targeting). The reason is that CAADTM gives higher scores to those POIs that a user prefers, that is, these types of POIs belong to the user's favorite content table. Then, CAADTM ranks POIs according to the corresponding giving scores. The traditional ad targeting does not operate favorite content table, so the hit rate of favorite advertisements of CAADTM is always equal to or greater than that of the traditional ad targeting method.

image

Figure 19. Comparison of favorite advertisement's hit rate between context-aware ad targeting method (CAADTM) and traditional ad targeting method.

Download figure to PowerPoint

6.2.2.3 Part2: comparison of the number of annoying advertisements after processing our proposed context-aware ad targeting method and traditional ad targeting method

This experiment mainly compares the number of annoying advertisements between our proposed CAADTM and traditional ad targeting method. We did this experiment five times. In Figure 20, there is not any annoying advertisement after processing CAADTM. The reason is that CAADTM gives the negative scores to those types of POIs, which are in the annoying content table filtering. Then, CAADTM deletes POIs, which have the negative scores from the list, that is, CAADTM does not push these advertisements to a user. The traditional ad targeting method does not consider the annoying content table of a user, so it is possible to push those advertisements that a user does not like.

image

Figure 20. Comparison of the number of annoying advertisements between context-aware ad targeting method (CAADTM) and traditional ad targeting method.

Download figure to PowerPoint

6.2.2.4 Part3: comparison of clicking advertisement feedback after processing our proposed context-aware ad targeting method and traditional ad targeting method

In this experiment, we show that comparison of clicking feedback between our proposed CAADTM and traditional ad targeting method. We click some types of advertisements for several times, so CAADTM may send these advertisements to a user. In Figure 21, the hit rate is the ratio of mapping between those types of the top five pushed advertisements and those advertisements that are clicked by a user. It is obvious to observe that the hit rate of processing CAADTM's advertisement clicking feedback is higher than that of the traditional ad targeting. The reason is that the CAADTM computes the weight of those types of POIs according to the advertisement clicking feedback. Hence, rank of those POIs is increased.

image

Figure 21. Comparison of clicking advertisement feedback after processing our proposed context-aware ad targeting method (CAADTM) and traditional ad targeting method.

Download figure to PowerPoint

From our observation, the CAADTM without the feedback technology provides a 56% hit rate of the pushed advertisement in average, which is illustrated in Figure 19. The CAADTM with the feedback technology provides a 68% hit rate of the pushed advertisement in average, which is depicted in Figure 21. Comparing with part 1 and part 3 depicted in this subsection, it shows that the proposed CAADTM method with the feedback technology provides a higher hit rate of the pushed advertisement than others.

7 CONCLUSION

  1. Top of page
  2. SUMMARY
  3. 1 INTRODUCTION
  4. 2 RELATED WORK
  5. 3 PRINCIPLE OF FINGERPRINTING-BASED POSITIONING METHOD
  6. 4 THE PROPOSED APPROACHING DETECTION METHOD
  7. 5 THE CONTEXT-AWARE AD TARGETING METHOD
  8. 6 EXPERIMENTAL RESULTS
  9. 7 CONCLUSION
  10. ACKNOWLEDGEMENTS
  11. REFERENCES

In this paper, we have proposed the PCA-MAS that includes two main parts: (i) ADM, and (ii) CAADTM. PCA-MAS can be used in an interested area, for example, Central and West District of Tainan City, for the mobile advertisement over cellular networks. PCA-MAS not only finds some POIs that a user is approaching but also finds and sends appropriate advertisements to a user/traveler. For the business model, advertisement operators can adopt the platform of PCA-MAS to charge stores if some feedbacks are sent back from users/travelers. To verify ADM and CAADTM, we have the case study of implementing PCA-MAS in Central and West District of Tainan City in the android platform over GSM networks. The proposed ADM and CAADTM also can be used in other cellular networks, that is, 3 G/3.5 G and 4 G.

The ADM mainly determines some POIs that a user is approaching using the concept of the fingerprinting-based positioning method. Some advantages of ADM are as follows: (i) lowering the cost to establish a fingerprinting-based database for positioning, (ii) using fewer records in the database to increase the searching efficiency for positioning, and (iii) being able to be applied in an region-specific advertisement system.

The other part of PCA-MAS is the CAADTM, which mainly pushes appropriate advertisements to a user. CAADTM takes some parameters as the concern. These parameters contain user's profile, current time, personal historic behavior, for example, the rate of clicking some types of advertisements, and public historic behavior, for example, popular rate. The advantages of CAADTM are as follows: (i) reducing the spamming advertisements that are sent to a user and (ii) achieving the goal of personalization in the mobile advertisement.

Our experiments have shown that (i) ADM has the good hit rate to determine some POIs that a user is approaching within the 150-m radius of the approaching range, and (ii) CAADTM has the good hit rate of finding appropriate advertisements that a user prefers through the favorite content table filtering, the annoying content table filtering, and the advertisement clicking feedback.

ACKNOWLEDGEMENTS

  1. Top of page
  2. SUMMARY
  3. 1 INTRODUCTION
  4. 2 RELATED WORK
  5. 3 PRINCIPLE OF FINGERPRINTING-BASED POSITIONING METHOD
  6. 4 THE PROPOSED APPROACHING DETECTION METHOD
  7. 5 THE CONTEXT-AWARE AD TARGETING METHOD
  8. 6 EXPERIMENTAL RESULTS
  9. 7 CONCLUSION
  10. ACKNOWLEDGEMENTS
  11. REFERENCES

This research is supported by the National Science Council of the Republic of China Taiwan under the contract number NSC 100-2219-E-006-002 and Institute for Information Industry (III).

REFERENCES

  1. Top of page
  2. SUMMARY
  3. 1 INTRODUCTION
  4. 2 RELATED WORK
  5. 3 PRINCIPLE OF FINGERPRINTING-BASED POSITIONING METHOD
  6. 4 THE PROPOSED APPROACHING DETECTION METHOD
  7. 5 THE CONTEXT-AWARE AD TARGETING METHOD
  8. 6 EXPERIMENTAL RESULTS
  9. 7 CONCLUSION
  10. ACKNOWLEDGEMENTS
  11. REFERENCES
  • 1
    Huang C-M, Yang C-C, Tseng C-Y. A telematics service discovery platform using the distributed peer-to-peer networking system. Software: Practice and Experience 2012. DOI: 10.1002/spe.2125.
  • 2
    Krumm J. Ubiquitous advertising: The killer application for the 21st century. IEEE pervasive computing 2011; 10(1): 6673. DOI: 10.1109/MPRV.2010.21.
  • 3
  • 4
  • 5
  • 6
    Niculescu D, Nath B. Ad hoc positioning system (APS) using AOA. The 22nd Annual Joint Conference of the IEEE Computer and Communications 2003; 7341743. DOI: 10.1109/INFCOM.2003.1209196.
  • 7
    Simsim MT, Khan NM, Ramer R, Rapajic PB. Time of arrival statistics in cellular environments. IEEE 63rd Vehicular Technology Conference 2006; 26662670. DOI: 10.1109/VETECS.2006.1683352.
  • 8
    Gustafsson F, Gunnarsson F. Positioning using time-difference of arrival measurements. IEEE International Conference on Acoustics, Speech, and Signal Processing 2003; VI – 553556. DOI: 10.1109/ICASSP.2003.1201741.
  • 9
    Charitanetra S, Noppanakeepong, S. Mobile positioning location using E-OTD method for GSM network. Student Conference on Research and Development 2003; 319324. DOI: 10.1109/SCORED.2003.1459714.
  • 10
    Spirito M, Mattiolli A. Preliminary experimental results of a GSM mobile phones positioning system based on timing advance. IEEE Vehicular Technology Conference 1999; 20722076. DOI: 10.1109/VETECF.1999.797302.
  • 11
    Trevisani E, Vitaletti A. Cell-ID location technique, limits and benefits: An experimental study 6th IEEE workshop on mobile computing systems and applications. 2004; 5160. DOI: 10.1109/MCSA.2004.9.
  • 12
    Takenga C, Kyamakya K. A Low-cost fingerprint positioning system in cellular networks. The 2nd International Conference on Communications and Networking in China 2007; 915920. DOI: 10.1109/CHINACOM.2007.4469533.
  • 13
    Huang CM, Hsieh TH, Lin SY. A Signal-aware fingerprinting-based positioning technique in cellular networks. The 14th International Conference on Network-Based Information Systems 2011; 325332. DOI: 10.1109/NBiS.2011.56.
  • 14
    Halim AH, Fauzi AH. Tarmizi and Selviawati. Bluetooth mobile advertising system using pull-based approach. International Symposium on Information Technology 2008; (4): 14. DOI: 10.1109/ITSIM.2008.4631881.
  • 15
    Lee JW, Lee CS, Park YS. Research on the advertisement effect of push type mobile advertisement. Fourth International Conference on Cooperation and Promotion of Information Resources in Science and Technology 2009; 137142. DOI: 10.1109/COINFO.2009.27.
  • 16
    Aalto L, Göthlin N, Korhonen J, Ojala T. Bluetooth and WAP push based location-aware mobile advertising system. Second International Conference on Mobile Systems, Application and Services 2004; 4958. DOI: 10.1145/990064.990073.
  • 17
    Kim YS, Lee JW, Park SR, Choi BC. Mobile advertisement system using data push scheduling based on user preference. Wireless Telecommunications Symposium 2009; 280284. DOI: 10.1109/WTS.2009.5068977.
  • 18
    Xu DJ, Liao SS, Li Q. Combining empirical experimentation and modeling techniques: A design research approach for personalized mobile advertising applications. Decision Support Systems 2008; 44(3): 710724. DOI: 10.1016/j.dss.2007.10.002.
  • 19
    Gao JZ, Ji A. SmartMobile-AD: An intelligent mobile advertising system. The 3rd International Conference on Grid and Pervasive Computing Workshops 2008; 164171. DOI: 10.1109/GPC.WORKSHOPS.2008.16.
  • 20
    de Castro JE, Shimakawa H. Mobile advertisement system utilizing user's contextual information. 7th International Conference on Mobile Data Management 2006; 91. DOI: 10.1109/MDM.2006.105.
  • 21
    Liapis D, Vassilaras S, Yovanof GS. Implementing a low-cost, personalized and location based service for delivering advertisements to mobile users. 3rd International Symposium on Wireless Pervasive computing 2008; 133137. DOI: 10.1109/ISWPC.2008.4556182.
  • 22
    Kuo1 S-P, Lin S-C, Wu1 B-J, Tseng Y-C, Shen C-C. GeoAds: A middleware architecture for music service with location-aware advertisement. IEEE International Conference on Mobile Adhoc and Sensor Systems 2007; 13. DOI: 10.1109/MOBHOC.2007.4428699.
  • 23
    Rashid O, Coulton P, Edwards R. Implementing location based information/advertising for existing mobile phone users in indoor/urban environments. International Conference on Mobile Business 2005; 377383. DOI: 10.1109/ICMB.2005.45.
  • 24
    Comput. Sci. Dept., Univ. Fed. de Campina Grande and Brazil. A context-aware system based on service-oriented architecture. Advanced Information Networking and Applications 2006; 1. DOI: 10.1109/AINA.2006.16.