دانلود مقاله ISI انگلیسی شماره 22637
عنوان فارسی مقاله

روش توصیه محصول: فیلتر کردن همکاری از طریق ارزش طول عمر مشتری و خواسته های مشتری

کد مقاله سال انتشار مقاله انگلیسی ترجمه فارسی تعداد کلمات
22637 2008 11 صفحه PDF سفارش دهید محاسبه نشده
خرید مقاله
پس از پرداخت، فوراً می توانید مقاله را دانلود فرمایید.
عنوان انگلیسی
Product recommendation approaches: Collaborative filtering via customer lifetime value and customer demands
منبع

Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)

Journal : Expert Systems with Applications, Volume 35, Issues 1–2, July–August 2008, Pages 350–360

کلمات کلیدی
سیستم توصیه گر - فیلتر کردن همکاری - فیلتر کردن بر اساس محتوا
پیش نمایش مقاله
پیش نمایش مقاله روش توصیه محصول: فیلتر کردن همکاری از طریق ارزش طول عمر مشتری و خواسته های مشتری

چکیده انگلیسی

Recommender systems are techniques that allow companies to develop one-to-one marketing strategies and provide support in connecting with customers for e-commerce. There exist various recommendation techniques, including collaborative filtering (CF), content-based filtering, WRFM-based method, and hybrid methods. The CF method generally utilizes past purchasing preferences to determine recommendations to a target customer based on the opinions of other similar customers. The WRFM-based method makes recommendations based on weighted customer lifetime value – Recency, Frequency and Monetary. This work proposes to use customer demands derived from frequently purchased products in each industry as valuable information for making recommendations. Different from conventional CF techniques, this work uses extended preferences derived by combining customer demands and past purchasing preferences to identify similar customers. Accordingly, this work proposes several hybrid recommendation approaches that combine collaborative filtering, WRFM-based method, and extended preferences. The proposed approaches further utilize customer demands to adjust the ranking of recommended products to improve recommendation quality. The experimental results show that the proposed methods perform better than several other recommendation methods.

مقدمه انگلیسی

Recommender systems have emerged in e-commerce applications to support product recommendation (Kim et al., 2001, Schafer et al., 2001 and Zeng et al., 2004), which provide individual marketing decisions for each customer. They assist businesses in implementing one-to-one marketing strategies, relying on customer purchase history to reveal customer preferences and identify products that customers may purchase. One-to-one marketing introduces a fundamental new basis for competition in the marketplace by enabling organizations to differentiate based on customers rather than products (Peppers & Rogers, 1993). Schafer et al. (2001) presented a detailed taxonomy of e-commerce recommender systems, and elucidated how they can provide personalization to establish customer loyalty. Generally, such systems offer several advantages, including increasing the probability of cross-selling, establishing customer loyalty, and fulfilling customer needs by presenting products of possible interest to them. Various recommendation methods have been proposed. The collaborative filtering (CF) method has been successfully used in various applications. It predicts user preferences for items in a word-of-mouth manner. User preferences are predicted by considering the opinions (in the form of preference ratings) of other “like-minded” users. The GroupLens system (Resnick, Iacovou, Suchak, Bergstrom, & Riedl, 1994) applied the CF method to recommend Usenet News and movies. Video recommender (Hill, Stead, Rosenstein, & Furnas, 1995) also used CF to generate recommendations on movies. Examples of music recommender systems are Ringo (Shardanand & Maes, 1995) and MRS (Chen & Chen, 2001). Siteseer Rucker and Polanco (1997) provided Web page recommendations based on bookmarks of user’s virtual neighbors, Amazon.com uses collaborative filtering to create books recommendations for customers (Linden, Smith, & York, 2003). Collaborative filtering requires a user to rate a reasonably large set of items, or the CF method has difficulty providing recommendations to novices (new users). Moreover, the CF method may suffer the sparsity problem, a situation in which transactional data is sparse and insufficient to identify similarities in user interests (Sarwar, Karypis, Konstan, & Riedl, 2000). Firms increasingly recognize the importance of customer lifetime value (CLV) (Berger & Nasr, 1998). Generally, RFM (Recency, Frequency, and Monetary) method has been used to measure CLV (Kahan, 1998 and Miglautsch, 2000). Identifying CLV or loyalty ranking of customer segments is important for helping decision-makers target markets more clearly in fiercely competitive environments. Additionally, the effect of CLV on recommendations should be investigated to make more effective marketing strategies. Recently, a weighted RFM-based CF method (WRFM-based CF method) (Liu & Shih, 2005b) has been proposed that integrates analytic hierarchy process (AHP) (Saaty, 1994) and data mining to recommend products based on customer lifetime value. This method employs association rule mining to identify recommendation rules from customer groups that are clustered according to weighted RFM values. Their experimental result demonstrated that the WRFM-based CF method can identify effective rules for making recommendations to customers with high lifetime value or loyalty. The WRFM-based CF method also suffers the sparsity problem. The content-based filtering (CBF) offers a different approach to collaborative filtering and provides recommendations by matching customer profiles (e.g., interests) with content features (e.g., product attributes). Each customer profile is derived by analyzing the content features of products purchased by the customer. The simplest of these techniques is keyword matching (Claypool, Gokhale, & Miranda, 1999). Krakotoa Chronicle (Kamba, Bharat, & Albers, 1995) is an example of such system. However, the CBF method is limited in not being able to provide serendipitous recommendations, because the recommendation is based solely on the content features of products purchased by the customer. Some domains, such as music recommendations, have difficulty analyzing content features of products. Several researchers are exploring hybrid methods of combining CF and CBF methods to smooth out the disadvantages of each (Basu et al., 1998, Claypool et al., 1999 and Good et al., 1999). This work uses customer demands derived from the frequently purchased products in each industry as valuable information to integrate the CF method for making recommendations. Extended preferences derived by combining customer demands and past purchasing preferences are used to alleviate the sparsity problem of recommendation. Different from conventional CF techniques, this work uses extended preferences to identify similar customers. Accordingly, this work proposes several hybrid recommendation approaches that combine collaborative filtering, WRFM-based method, and extended preferences. Moreover, customer demands are considered in re-ranking recommended products to improve the quality of recommendation. The remainder of this paper is organized as follows. Section 2 reviews related works on the typical KNN-based CF method, the WRFM-based method, hybrid works, and content-based filtering methods. Next, Section 3 outlines the proposed methods. Section 4 then describes the experimental setup and criteria to evaluate recommendation quality. Experimental results are also presented to confirm differences between methods. Finally, Section 5 draws conclusions, summarizing the contributions of this work and outlining areas for further research.

نتیجه گیری انگلیسی

The collaborative filtering method has been successfully used in a number of applications, but suffers several limitations. This work uses customer demands derived from frequently purchased products in each industry to integrate with the CF method to make recommendations. This work also combines customer demands and past purchase preferences to reduce the sparsity of customer-item matrix and further improves recommendation accuracy. Customer demands are included as a factor in re-ranking candidate products to provide recommendations. Several experiments were conducted to compare the effectiveness between various methods. According to the experimental results, generally, the performance ranking of those methods with extended preferences is WRFMEP ≻ CFEP ≻ EP-based k-NN method; while the ranking of those methods without considering extended preferences is WRFMCP method ≻ WRFM-based CF method ≻ preference-based CF method ≻ KNN-based method. This ranking implies that extended preferences, derived by combining customer demands and purchase preferences, are useful for improved recommendation quality. Furthermore, re-ranking candidate products according to customer demands offers a promising approach to improve recommendation accuracy. Finally, the experimental results show that the proposed hybrid methods not only improve the overall quality of recommendation, but also can be extended to recommend product items to customers who purchased few product items based on extended preferences. In general, the quality of recommendation improves as the number of purchased items increases. Future works will address three themes. First, the proposed approach was evaluated experimentally using a data set obtained from a hardware retailer. Further studies are needed to evaluate the application of the proposed approach to other application domains. Second, the present work focused on product recommendation of retail transaction data which contains binary choice of shopping basket data; the customer preference is represented as one, if the customer purchased the product; and zero, otherwise. Further investigation is needed to evaluate the effectiveness of the proposed methods for data sets with non-binary preference ratings. Finally, owing to the limitations of available content information of the data set concerned, this work could not address new and unseen items. Further studies are required to verify the proposed methods on other real cases that can support more content information.

خرید مقاله
پس از پرداخت، فوراً می توانید مقاله را دانلود فرمایید.