Searchdriven memorybased collaborative filtering for small and medium

Search-driven memory-based collaborative filtering for small and medium scale B 2 C e-Commerce UDO, Ifiok James; AFOLABI, Babajide; JAMES, Idara Paper presentation @ Conference on Transition from Observation to Information to Knowledge University of Lagos, Lagos 25 -26 th August, 2016

Presentation outline • • • Introduction Statement of the problem Aim & objective Related work Methodology Proposed Architecture Similarity measurements Results and evaluation Conclusion 2

Introduction • Memory-based CF: CF is a technique of making automatic recommendations about the interests of a user by collecting items preferences or taste information from other users (Santhini et al. 2015) • Search: Search is a technique used to retrieve information from its databased on keywords. • Small and medium scale B 2 C e-commerce: B 2 C e-commerce with inadequate capital. 3

Statement of the Problem • Memory-based CF approach adopts rating on item-toitem and user-to-user preferences to recommend items for the users. • Also, it is cost efficient in terms of implementation. • Nevertheless, lack or inadequate reviews of items via ratings by prior users limits recommendation. • Therefore, in order to increase awareness of the active users of the unknown items to prior users via recommendation; a search-driven rating is implemented on memory-based CF. 4

Aims of the work • Recommendation of items that may not have been rated by prior users. • Accurate recommendation of items to prior users 5

• Related Works Memory-based Collaborative Filtering (CF) o. Items rating (Wang et al. 2006) • Positive and negative ratings o. Similar users and preferred items • Similarity measures (such as Pearson correlation, Cosine similarity, Euclidean distance, Vector Space Similarity) o. The limitation • lack of initial rating data needed to make relevant recommendations due to inadequate reviews. • Other available CF models are model-based and hybrid approaches (Abdullah, 2012). oexpensive in terms of implementation 6

Our Approach Figure 1: Search-driven memory-based CF Architecture 7

METHODOLOGY • Jaccard similarity measure o Weighted similarity measurement for Retrieved Items (Equation 1) o Weighted similarity measurement for user and items (Equation 2) o Scoring Items with Users’ Rating • average positive rating (Equation 3) • average negative rating (Equation 4) 8

back 9

Summary of Retrieved Results Relevant Items Phones Non-relevant User-Item Search-driven 1 7 5 6 8 2 23 22 1 8 7 5 14 7 18 20 5 8 3 6 6 3 28 45 Retrieved Non-retrieved Shirts/ blouse Retrieved Non-retrieved Laptops Retrieved Non-retrieved Table 1: Summary of the recommendation results obtained from user-item rating and the search-driven approaches 10

Evaluation Figure 2 a Figure 2 b Figure 2 a and 2 b: Graphs showing the Recall and Precision for the items in user-item and search-driven approaches of memory-based CF 11

CONCLUSION • Search-driven memory-based CF approach presents more products recommendation than user-item rating. • Products unrated by prior users are included in the recommended results for active users. • Therefore, could be deployed by small and medium scale B 2 C e-commerce. 12

List of References • Abdullah, N. (2012). Integrating Collaborative Filtering and Matchingbased Search for Product Recommendation. Ph. D Thesis School of Electrical Engineering and Computer Science, Queensland University of Technology. • Santhini, M. , Balamurugan, M. and Govindaraj, M. (2015). Collaborative Filtering Approach for Big Data Applications based on Clustering. International Journal of Mathematics Computer Science and Information Technology, Vol. 2 Issue 1, pp 202 -208. • Wang J. , Vries, A. P. and Reinders, M. J. (2006). Unifying user-based and Item-based Collaborative Filtering Approaches by Similarity Fusion. In proceedings of the 29 th annual international ACM SIGIR conference on Research and development in information retrieval, pages 501 -508, USA. 13

THANK YOU 14