A multidimensional method for context-aware bidders in mobile commerce

Master's Thesis in Computer Science, Software Orientation

The use of context, as dynamic information that describes the status of users and items and affects the decision-making and selection process of users, is necessary by the recommender systems in mobile commerce in order to improve the quality of the offer. In this research, a new multidimensional method for context-aware bidding in mobile commerce is presented. In this method, users' information, items, context parameters and the relationship between them are displayed in a multidimensional space, which is called a multidimensional score cube. In this space, similar fields are identified separately for each user, which is done by identifying different consumption patterns of users in different field conditions. By obtaining this information, a new two-dimensional space is created and the final proposal is made using a collaborative filtering method in this space. The evaluation of the method through its implementation in a restaurant food products suggestion system including the parameters of the day, time, weather and accompanying fields in addition to the parameters of the user and items and comparing it with the traditional suggestion method without considering the background information has been done. Self-organizing networks have been used to implement the cooperative filtering method. Self-organized networks are a type of unsupervised neural networks. The comparison and evaluation of the results has been done by calculating the F1 index, which is one of the standard and widely used indicators for the evaluation of bidders. Based on these results, the multidimensional bidding method shows an improvement of about sixteen percent compared to the traditional bidding method, which confirms the effectiveness of the method in terms of the quality of bidding. Keywords: context-aware bidding systems, mobile commerce, self-organizing map, multidimensional bidding method, collaborative filtering. Chapter 1: Introduction The bidding systems in mobile commerce have been among the most important topics in recent years, with The emergence of wireless technologies and facilitating the movement of e-commerce from wired to wireless environments have been considered. Mobile commerce means carrying out e-commerce activities through wireless environments, especially wireless internet, and mobile handheld devices, with the emergence of wireless technology in the field of the Internet and the increasing use of mobile devices, attention is increasing [1, 2]. Two special characteristics of mobility [1] and wide access [2] have been attributed to mobile business applications [1,3], the first characteristic emphasizes the possibility of the disappearance of location restrictions and the second characteristic emphasizes the possibility of the disappearance of time restrictions in users' use of the services of these types of applications [1,3,4,5]. The fact that users are able to replace devices such as mobile phones and personal digital assistants (PDAs) [3] instead of personal computers to carry out activities such as electronic banking or electronic purchase of products will provide many facilities for them and new opportunities for businesses and shows the need for researchers to pay attention to this field [1,3]. But the implementation of recommender systems in mobile environments without It will not be appropriate to consider the influencing parameters in this environment. The set of these parameters constitute background information [6]. These resources can include items such as specific information required by the user or products such as a book or movie that a user likes among the multitude of products that the user is facing [7, 8, 9]. In recommender systems, there are three main data sets, that is, the set of users (C), the set of recommendable items (S) (such as books, movies, music, etc.) and the sets of data that define the relationship between the two previous sets. Collection S can contain hundreds, thousands and even millions of goods in different applications, and similarly collection C can have such a situation. The relationship between two sets C and S is based on a scoring structure that determines the usefulness or interest of the product for the user.This relationship with a function called the utility function, u, is defined as the following relationship [7]:

(1-1)

where Ratings is an ordered set such as non-negative integers or a set of real numbers in certain spaces.

In the systems that recommend the values ??of u, it is usually only defined on subsets of the C×S domain and not on all of it, and the uncertain parts of this domain should be estimated using the available data. specified The ultimate goal of recommender systems is achieved by offering the items with the highest estimated scores to users, so that for each user, the items with the maximum amount of usefulness are selected and introduced [7].

To date, many suggestion methods have been presented, and these methods and methodologies fall into the following categories [7,9,10]:

Content-based [4]: ??In this group of methods, the act of suggesting using Finding items that are most similar to items that have been the user's favorite in the past. In other words, u(c, s), the usefulness of product s for user c is estimated based on all available values ??of u(c, si) where si is similar to s and si are among the user's favorite products.

Participatory filtering: In this group of methods, the act of offering is done by finding items that are liked by users with similar tastes. Users with similar tastes means users who rated the same items similarly. In other words, u (c, s) is obtained based on the available values ??of u(cj, s) that cj users are similar to c. Hybrid model [5]: methods that combine two methods based on content and collaborative filtering and in this way take advantage of the advantages of both methods in order to identify and introduce goods. In another view, the methods of offering, both content-based and collaborative filtering, are divided into two categories. Memory-based [6] and model-based [7] methods are divided. In comparison with memory-based algorithms, model-based algorithms, using machine learning methods [8], create a model using the set of existing points and use it to predict points [7,10,11]. With the emergence of wireless technology in the field of the Internet and the increasing use of mobile devices, the implementation of recommender systems in mobile environments requires further investigation in order to provide more relevant and personalized information in view of its specific limitations such as the high cost of connection time and data exchange, bandwidth limitations, low quality of connection and the limitations of the input and output of mobile devices. Investigating the impact of background information as the user's conditions and environment and as information that affects his decision-making process, on the output of such applications, are the issues that have been investigated in this research.

Research background

The emergence of wireless technologies and the increasing use of mobile devices, have created many opportunities for electronic commerce applications. Considering the specific limitations of mobile environments, providing information in a more personalized and customized manner is one of the important goals of mobile business applications. Considering background information as the user's conditions and environment and as information that affects his decision-making process, in providing the output of such applications is one of the things that can be used to provide more relevant information to users.

Recommender systems have always been among the most important issues in the field of e-commerce. Context-aware mobile recommender systems are just getting started. An important category of context-aware systems are location-aware systems. Yang, Cheng, and Daya[12] present a location-aware recommender system for mobile environments, which aims to recommend sellers' websites by taking into account the customer's interests and preferences, as well as his location distance from the physical location specified on the websites. In the mentioned method, the above two factors are calculated separately and then websites are suggested based on their combination. Another such system is Proximo[13], which is a location-aware recommender system for indoor environments such as museums and galleries. This system suggests items based on the user's interests and defaults and displays the location of the items on the roles on the user's mobile device.