WWW 2007 / Poster Paper

Topic: Services

A Novel Collaborative Filtering-Based Framework for Personalized Services in M-Commerce
Qiudan Li, Chunheng Wang, Guanggang Geng, Ruwei Dai
Key Laboratory of Complex Systems and Intelligence Science, Institute of Automation, Chinese Academy of Sciences, Beijing

{qiudan.li, chunheng.wang, guanggang.geng, ruwei.dai}@ia.ac.cn

ABSTRACT
With the rapid growth of wireless technologies and handheld devices, m-commerce is becoming a promising research area. Personalization is especially important to the success of mcommerce. This paper proposes a novel collaborative filteringbased framework for personalized services in m-commerce. The framework extends our previous work by using Online Analytical Processing (OLAP) to represent the relations among user, content and context information, and adopting a multi-dimensional collaborative filtering model to perform inference. It provides a powerful and well-founded mechanism to personalization for mcommerce. We implemented it in an existing m-commerce platform, and experimental results demonstrate its feasibility and correctness.

Based on the excellent framework of m-commerce in [1] and the technique of multi-dimensional collaborative filtering in ecommerce [3], we propose a novel collaborative filtering-based framework for personalized services in m-commerce, which develops and enriches our previous work [2]. The basic idea is to first identify the nearest neighbors of an active mobile user by analyzing the similarity relations between the user and the history mobile users under some specific context information, and then perform personalized services based on the similar neighbors found above. Within this framework, OLAP is used to represent the relations among user, content and context information, and a multi-dimensional collaborative filtering model is adopted to perform inference. The framework can take full advantages of the multi-dimensional representation ability of OLAP and the function of collaborative filtering.

Categories and Subject Descriptors
H.2.8 [Database Applications]: Data mining; H.3.5 [Online Information Services]: Web-based services.

2. COLLABORATIVE FILTERING-BASED FRAMEWORK
Figure 1 shows the flow chart of the proposed framework which consists of three components. They are, (1) to collect user, content and context information; (2) to represent the multidimensional information by OLAP; (3) to perform inference via collaborative filtering. We introduce them in the following sections.

General Terms
Algorithms, Experimentation, Human Factors.

Keywords
Personalization, collaborative filtering, m-commerce.

1. INTRODUCTION
With the rapid growth of wireless technologies and handheld devices, mobile commerce (m-commerce) is becoming a promising research area [1]. It is defined as the transactions of commodities, services, or information over the Internet through the use of mobile handheld devices. However, due to the inherent limitations of mobile devices and wireless network, a key factor for effective m-commerce is to deliver relevant information to the right people at the right time in the right way. Hence, personalization is especially important to the success of mcommerce. Providing personalized services is about a process of matching users and merchants based on their profiles and preferences in conjunction with a changing environment constituting several context factors such as time, location and weather. In previous work, we have identified that the matching process requires effective representations of the information (including user, content and context) relationships and the inferring abilities based on the representations [2].
Copyright is held by the author/owner(s). WWW 2007, May 8-12, 2007, Banff, Alberta, Canada. ACM 978-1-59593-654-7/07/0005.

Figure 1: Flow chart of the proposed framework

2.1 Source Collection Layer and Relationship Presentation Layer
Survey technique is employed to gather the user information such as demographics (income range, ethnicity, work experience, etc.) and preferences. Context information can be obtained as follows: mobile user's location can be obtained through the PDAembedded GPS module (outdoor) or mobile positioning system

1251


WWW 2007 / Poster Paper (indoor); the time is provided by the portable device; the weather is obtained from the website of the online weather services; and user's current activities can be identified either from his/her schedules, or by mining patterns from his/her historical data. Content information is mainly uploaded by merchants through web-based merchant interfaces or collected by mobile agents from merchants' websites. Inspired by the idea of the paper [3], we redefine the recommendation space of m-commerce by OLAP as follows, where the information mentioned above is considered. Definition1 Recommendation Space: Recommendation space, consisting of user, content items, context (Location, Time, Weather, Activity) is represented as: User × Item × Location × Time × Weather × Activity . Accordingly, the ratings on the recommendation space are stored in a multidimensional cube.

Topic: Services The quality of recommendation is measured by "precision", which is widely used in information retrieval area. The definition is given as follows [4]:

Pr ecision =

size of hit set size of Top - N set

Based on the metric, we compute the average precisions on test set. The comparison result of the proposed framework and traditional collaborative filtering method (without considering context information) is shown in Figure 2.

2.2 Collaborative Filtering based Inference Layer
The multi-dimensional recommendation achieves a great success in e-commerce [3]. Encouraged by this achievement, we perform inference by the technique of the multi-dimensional recommendation for m-commerce. The inference process can be summarized as follows: Step1: For an active mobile user, identify his/her context information; Step2: Based on the context information, and OLAP-based multidimensional recommendation space, perform two relational operations: selection followed by projection, which obtains a twodimensional recommendation representation; Step3: This step involves two components, firstly, based on the produced two dimensional representation space User × Item , to find active user's nearest neighborhoods according to some similarity metric; secondly, to perform top-N recommendations based on the rating information of the found neighborhood, namely, scans through their rating items and calculates the frequency count for the restaurant items, then, sorts the items according to the count, and finally, produces the top-N items list, where the items are most frequently rated by the neighbors and have not yet been rated by the active user. Figure 2: Average precision comparisons of two methods From the results, we can find that the precision of our framework is higher than that of traditional collaborative filtering method. The results show that since our framework identifies the relations among user, content and context information efficiently, so the recommendation quality is improved.

4. CONCLUSIONS and FUTURE WORK
In this paper, we presented and evaluated the novel collaborative filtering-based framework for m-commerce applications, which is based on OLAP and the multi-dimensional collaborative filtering. Experimental results show that our framework can produce more accurate personalized services for m-commerce. Building more effective models to explore latent relations among neighbors is still difficult and remains the subject of ongoing research.

5. REFERENCES
[1] Varshney, U., Vetter, R.: Mobile Commerce: Framework,
Applications and Networking Support, Mobile Networks and Applications, Vol.7, (2002) 185-198.

3. EXPERIMENTAL RESULT
We implement the proposed framework in an existing mcommerce platform of food industry to provide Top-N recommendation, and compare its recommendation quality against that of traditional collaborative filtering. The traditional collaborative filtering represents the recommendation space by a two dimensional user-item matrix and does not consider context information. We use cosine-based function as the similarity metric and N = 5 as the maximum number of items, which will be recommended by our new method. In our experiment, we test different methods on the data set, which comes from user's rating data stored in Data Service Layer of the m-commerce platform. The data collection describes users' preferences on various restaurants under different context conditions. We randomly choose 200 users and 30 restaurants items. The data set is divided into 80% training set and 20% test set.

[2] Liao, S., Li,Q.D., Xu, J.J.: A Bayesian Network-Based
Framework for Personalization in Mobile Commerce Applications. Comm.AIS, Vol.15, (2005) 494-511.

[3] Adomavicius, G., Sankaranarayanan, R., Sen., S., Tuzhilin,
A.: Incorporating Contextual Information in Recommender Systems Using a Multidimensional Approach. ACM Transactions on Information Systems, Vol.23, (2005) 103145.

[4] Sarwar, B., Karypis, G., Konstan, J., Riedl, J.: Analysis of
Recommendation Algorithms for E-Commerce. In Proceedings of ACM E-Commerce, (2000) 158-167.

1252