R ECOMMENDATION TYPE

國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

12

2.2 Recommendation type

With the advances in information technology and the used in the Internet have accelerated the diversities and various alternative items in different domain in online, like thousands of songs, restaurant and hotel etc (Burke 2002). When the collections are increasing and information overloading, individual is force to time consuming on choosing the item. Recommender systems are design for solving these problems with friends and experts/mavens who have the knowledge about the item/product. Recommender systems most applied on e-commerce service in several categories (Nenkova, 2012). In general point view, Recommender system are help for suggesting the suitable items that user are interesting, meanwhile, benefit both the user and the item provider.

Recommender systems have been an important application and start forcing of considerable recent academic like customer behavior area (Nenkova, 2012).

Recommender systems utilize several technological methods to archive the recommend task. Most of the commonly technologies used in Recommender systems today can generally categorized into content-based filtering, collaborative filtering, demographic filtering, hybrid recommender systems, trust based social recommender systems, agent based recommender systems (Nenkova, 2012). We reviewed all of these technologies and compare each other, for find out the most suitable technologies for impact on customer engagement behavior.

Content-based filtering method utilizes the description or content of items to filtering and recommend. It used to item-to-item relationship to bridge the user needs and performance. In order to enhance the direction on recommendation on item, content-based system processes information from

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

13

various resources and needs to extract the useful characteristics and elements about its content. Each recommender would define its own description and element to facilitate the relevance of items. The advantage of content-based filtering is no need of the user historical data. Without using the user rating, they are able to recommend new item and unpopular item with less user rating to customer. On the other hand, the disadvantage is difficulty to analyze the multimedia content and finding something unexpected without any categorize (Burke 2002).

Collaborative filtering uses the collections of the rating on a list of items for user suggestion. Opinions can be allowedly given by the user as a rating score or can be explore from the historical data of the user. Most of the Collaborative filtering divide into two categories: User-based and Item-based.

User-based cumulate the correlation with other user and collecting the scoring information from user. With this method, user-based collaborative filtering predicts the scores of the unrated items according to the historical rating data; the item-based collaborative filtering predicts the score of the item by averaging the current user’s rating data of similar items in the past, item with sparse data is less important,. With this matrix, the item-based cause fewer problems with cold start and attacks comparing with the user-based. Generally, collaborative filtering does not need a representation of items of feature, only based on the participant and involvement of user community. This brings the customer behavior inside the chosen, but still conflict in different individual with different thought (Nenkova, 2012).

Demographic filtering is to build models by other past user with clustering the stereotypes and the characteristics. A typecast is a collection of the characteristics and knowledge which frequently use by users or user groups. The

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

14

purpose is to quickly setting the new customer into a relative typecast, that can increase the directly of the recommend. The weakness of the demographic filtering is that model recommended based on clustering some similar characteristic, for example interest, the system might provide recommendation which is too general.

Hybrid recommender systems combine two or more technologies to gain better benefit with other advantage; this is another category of recommender systems to overcome the limitations of the other approach. Most common hybrid recommender is to combine content-based and collaborative filtering to enhance the relative of the recommend on content similarities and user rating perspective.

Hybrid recommender system including various types: Weighted, switching, Mixed, Feature combination, Cascade, Feature augmentation and Meta-level. To fulfill the dynamic e-commerce environment, hybrid recommender systems is suitable solution to adapt the environment needs (Göksedef, 2010).

Trust based social recommender systems integrate with the user community. The trust factor established in the user will be aware of the nature of the recommendations. Based on derived from information about user profiles and relationships between users, trust based social recommender systems emerges some rules that can be played by explanations in RSs, for example, trust , satisfaction, persuasiveness. In the Trust based social recommender systems, system provide recommendation based on the human interaction emotion and behavior (Nenkova, 2012).

Overall, after reviewing the several technological, we discover that the trust based social recommender systems establish by the human behavior versus the other technological with functional-oriented of few interactions. For the purpose to influence on user behavior, we expect that our research can have the

‧

stronger effect on the customer engagement behavior on the social recommender systems media.

Table 2.2 the overview of the recommender technologies

Technique Content-based Collaborative Filtering Demographic

Filtering engagement on maven’s behavior since the trust-based social recommender is recommended by user knowledge. Maven are described as ‘‘individuals who have information about many kinds of products, places to shop, and other facets of markets, and initiate discussions with consumers and respond to requests from consumers for market information” (Feick and Price, 1987) .However, the mavens are deeply involved in a wide range of categories and understand the information about the service or detail of product even they do not use (Boster, 2011). Maven's influence is based on more general market expertise, besides, mavens also like to discuss the deals they get and frequently volunteer advice to others regarding purchasing decisions. Typically mavens are recognized by others as such and are sought after by others for information (Boster, 2011).

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

16