In this section, future steps of this research are discussed. As mentioned earlier, the present research aims to study the main sources of error that affect the performance of a UBPrM. As the first step, a novel approach called Event Profiling Method (EPM), which benefits from the underlying similarity between users and the underlying patterns existing in the events, was proposed in order to reduce the impact of the ME. To implement and test EPM in a real-world scenario, we selected video popularity prediction as the first use-case.
One of the next major steps of this research is to verify and evaluate the assumptions and ideas of the EPM approach for other real-world scenarios. For this purpose, We propose the service demand and spatial dispersion prediction of the users of telecommunication network as the second use-case for EPM. According to this, this step aims to develop a model for predicting the situations in which a large group of users request a service in a specific location and a specific time. On the other side, as discussed in chapter 2, although exploiting social ties amongst users is able to improve any UBPrM, the current version of EPM is not designed to take advantage of this valuable information.
Then, we utilized the ideas of EPM to develop a novel approach for improving V-PPMs. A V-PPM has a wide range of applications ranging from caching techniques to broadcasting technologies to marketing industries. As the results, the next major step of this part of our research is to evaluate the impact of the proposed approach on the applications of V-PPM.
Afterwards, we developed a combinational technique to improve the user profiles to alleviate the impact of SaE on the performance of a user behavior predictive model. Despite of the presented improvements, the conditions of the proposed technique are not optimal and the technique does not consider the time dependencies of user behaviors. Therefore, two of the main future steps for this part of our research is to enhance the combinational technique such that the conditions have their optimal form and the technique capture the time dependencies as well. In addition, the impact of the proposed technique on a UBPrM needs to be tested.
Considering the above-discussed points, we can list the future steps of this research as follows.
• Future steps of EPM
– One of the factors that affects a user’s behavior is the behavior of other users who have a social relationship with the user. Our state-of-the-art review revealed that using information about the social ties of users considerably improves the predictive models. Furthermore, modeling and utilizing the social information based on existing techniques is computationally expensive. As the result, they
may not be suitable for applications that require fast predictions. In this regard, a future challenge for this research is to find a solution for efficient integration of social information into EPM.
– To justify the assumptions and ideas of EPM, we need to implement it in different real-world use cases. In addition to the popularity prediction, service demand and spatial dispersion modeling is proposed as another use-case. This use-case has important applications in network optimization and service personalization. To exploit the service demand and spatial dispersion as a use-case for EPM, one needs to develop a method based on EPM to predict the spatial dispersion of users and their aggregated service demand (which can be considered as an event). The developed method may benefit from the presented techniques for user grouping, dominant-follower identification and classification. The evaluation of the accuracy of the service demand and spatial dispersion method requires an empirical dataset. The dataset ought to contain the time and location (with minimum granularity of BTS) of the services that have been demanded by users.
• Future steps of the proposed V-PPM
– Caching contents in network nodes to put the contents near to clients is a key part of Content Centric Networking (CCN). The performance of the used caching scheme is important for an efficient content delivery time. Recently, a number of popularity-based content caching algorithms have been proposed to improve the efficiency of the content caching in CCN. As the result, the performance of the popularity-based caching algorithms depends on the accuracy of the employed C- PPM. In chapter 4, we showed that our proposed approach results in a significant improvement in the performance of V-PPMs. One of the remaining steps is to test the impact of the improved prediction results on the performance of the existing popularity-based caching schemes.
– Predicting the popularity of video contents is of great importance for broadcast TV operators. On one side, broadcasting the popular programs results in a reduction in the average time of finding the TV programs by users. This improves users’ satisfaction and retention. On the other side, a broadcast TV operator is able to manage and optimize the network resources in advance based on the prediction results of a popularity prediction model. According to the presented discussion, the efficiency of the broadcasting mechanisms is correlated to the accuracy of the utilized content popularity prediction model. The other future step for this research is to enhance the existing broadcasting mechanisms by
the proposed V-PPM and evaluate its impact on the efficiency of the enhanced mechanism.
– Targeted advertising is defined as an advertising in which advertisers target the most appropriate audiences. The set of the most appropriate audiences is found based on a number of certain traits of the audiences. The traits can be demographic or psycho-graphic. Examples of demographic traits include sex and age; and examples of psycho-graphic are personality and interests. As discussed in the previous section, by analyzing the results of our proposed V-PPM, one can obtain some demographic and psycho-graphic information about users. For illustration, analyzing the most popular contents in each user group provides information about sex, age and interest of users in the user group. According to this discussion, a future step for this research is to develop a method to estimate the demographic and psycho-graphic information about users. This step requires multi-disciplinary research that links user behavior modeling to psycho-analysis area.
• Future steps of the proposed combinational technique
As discussed in chapter 5, the present study is a proof-of-concept of practicality of the proposed combinational technique that just attempted to verify the effectiveness of the proposed technique. There are some future steps to improve this technique that are presented in the following.
– The conditions are selected such that the statistical accuracy is improved; but they are not the most optimal conditions that can maximize the improvement in the statistical accuracy. In finding the conditions, we weighted more on their practicality than their optimality. So, the next natural step of this study is to look for the most optimal form of the conditions and in the same time, a practical form of them.
– The proposed combinational technique is designed to improve the time indepen- dent statistics, such as probability of watching a programme, of user profiles. However, there are a number of useful time dependent statistics, such as users’ reaction time to different released videos and the popularity of the videos, for user profiling models. As the result, one of the future steps for this part of our research is to enhance the technique such that the technique is able to capture the time dependent statistics.
– It was shown that the proposed combinational technique in this thesis is able to reduce the SaE of user profiles. As the result, another future step of this
research is to enhance the existing user behaviors predictive models, specially our proposed V-PPM in chapter 4, by this method and evaluate the impact of this method on their performance. The proposed combinational technique can also be used to reduce the impact of SaE on the performance of the existing recommender systems, users’ activity and mobility prediction models. Therefore, the impact of the proposed technique on the performance of the mentioned predictive models needs to be tested, as well.