Guide and Retain Users
Zhang Min · 2018
With the rise of the Internet, especially the mobile Internet, massive information is presented on the Internet and users often get lost. Recommender system severs as mining valuable information presented to the users in the vast amount of information in social networking, news, video and other major mainstream applications. The current recommendation system mainly performs offline calculation of some intermediate data of the recommendation system, such as the similarity between the items, and then recommends the items that the users are most interested in when the users log in the application. Since data sparsity and cold-start problems always exist, the improvement effect is not obvious only improving recommended algorithm currently. Also the user's interest may be not single in real-time recommendation, the current recommender system presents the recommended results via the offline calculations solely and does not guide the user to discover more interests in real time. Therefore it's necessary to study the continuous recommendation of recommendation system from the interaction for two purposes:(1)Recommender constantly gets users' preferences to solve data sparsity and cold-start problems so as to recommend items to users which is more close to users' preferences;(2)Recommender guide users to dig users' more interests to avoid the case that recommender systems always recommend items based on the user's historical interests. Firstly, this paper proposes that the recommender allows the users to explore items which are similar to the items filtered by the conditions that the users specified. Secondly, the recommender let users offer ratings to items to get user's more interest to improve accuracy of recommended result through conversational collaboration model when users can also explore the similar items and rate it. Thirdly, in order to increase the transparency of the recommended system, we show the explanation list to users, and in order to increase the controllability of the recommended system, we allow users to change the intermediate data, and in order to improve the intuitiveness of the recommendation system, we visualized the recommended results. Lastly, we combine the critique-based recommender system and collaborative filtering recommender system to constantly revise the recommended results and users can decide the weight of the two recommendation methods and the weight of each critique.