Identifying the Most Influential Product with Cost Constraint

• Main Authors: Cheng-HungChen, 陳政宏
• Other Authors: Chiang Lee
• Format: Others
• Language: en_US
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/04528616779585597127

碩士 === 國立成功大學 === 資訊工程學系 === 102 === Top-k queries are widely used for retrieving the k most interesting data points (objects) based on user preferences. Consider a customer wants to buy a smart phone, but he/she does not know how to choose from a lot of smart phones. In this case, top-k query is used for retrieving the k most interesting products based on the user preferences. On the other hand, if a product is in the top-k list of a customer, this product may have a high opportunity to be purchased by this customer. This information is an important business opportunity to manufacturers. Thus, we can extend the concept of top-k query processing to find the reverse top-k query for the manufacturers. By using the reverse top-k query, the top-k list of customers can be derived to manufacturers. Furthermore, the manufacturers can figure out the preferences of customers. In this thesis, we will discuss how to identify the most influential product with cost constraint. We define the influence of product as the number of customers who choose this product. The previous researches [2,3] discussed how to find the influence score of the query points with the given products. However, applying these algorithms of [2,3] to our problem has to enumerate all the points on the cost plane. Obviously, it results a high computational cost and impractical. Therefore, we propose two algorithms, the cutting the cost plane algorithm (CCA) and the grid-tree based algorithm (GA), for identifying the most influential product with cost constraint. CCA evaluates each user preference to find the greatest influence score region and GA is based on grid-tree technique to find the answer efficiently. In addition, we designed a series of experiments to demonstrate and evaluate the efficiency of our proposed algorithm. Our experiments exhibit the proposed algorithms are practical and efficient in retrieving the most influential products.