结合用户间相互关注的个性化APP图推荐模型研究

Along with the extensive use of smart phones, lots of apps appear on the web. Facing those available apps, how to acquire the suitable apps for users has already become a hot issue researched by many famous web companies. However, until now, the generally acceptable way for app recommendation is first to calculate the matching degree between app’s context and user’s interest labels. Then, the apps with the larger matching degrees are recommended to user. Unfortunately, there are three problems occurring in this kind of recommending methods. 1, there are many duplicate apps in the recommending results; 2, the implicit interests of users cannot be detected; 3, the related apps cannot be recommended together. There are two reasons to the aforementioned problems. One is that it lacks of an effective way to calculate the relationships between apps. The other is that there is no effective means to acquire user’s interests without using user’s manual labels. For these two reasons, how to construct a graph based personalized app recommendation model which can satisfy user’s need is researched in this project. This model is formed based on the relationship calculating results between apps. Besides, the user’s implicit interests can be acquired by this model via importing user’s mutual attentions. These interests can help recommendation model recommend more suitable apps to users. Using the app data provided by Application Treasure of Tencent, the user’s behaviors (including hits, downloads, and reviews) can be used in the recommendation model to further improve recommending accuracy.

随着智能手机的广泛普及，手机应用（APP）如雨后春笋般涌现。面对数量如此众多的手机应用，如何让用户获得满意的应用已经成为各大门户网站竞相研究的热点。然而，迄今为止，普遍采用的应用推荐方法仍是首先计算应用和用户兴趣标签的匹配度，然后将匹配度大的应用推荐给用户。此种推荐方式有三个问题：1，推荐的应用重复度很大；2，无法发现用户隐含的兴趣爱好；3，无法将具有相关性的应用一同推荐给用户。上述问题存在的核心原因是缺乏一种有效的应用间相关度计算方法，同时缺乏有效的手段能够自动获取用户的兴趣爱好。针对以上问题，本项目研究了面向手机应用的个性化图推荐模型构建问题，该模型以应用间的相关度计算结果为基础，同时通过结合用户间的相互关注能够挖掘出用户隐含的兴趣点进而向用户推荐更加适合的应用。借助腾讯公司提供的应用宝数据，本项目还能够利用用户的行为数据（包括用户的点击、下载、评价行为）进一步提升推荐模型的准确度。

随着智能手机的广泛普及，手机应用（APP）如雨后春笋般涌现。面对数量如此众多的手机应用，如何让用户获得满意的应用已经成为各大门户网站竞相研究的热点。然而，迄今为止，普遍采用的应用推荐方法仍是首先计算应用和用户兴趣标签的匹配度，然后将匹配度大的应用推荐给用户。此种推荐方式有三个问题：1，推荐的应用重复度很大；2，无法发现用户隐含的兴趣爱好；3，无法将具有相关性的应用一同推荐给用户。上述问题存在的核心原因是缺乏一种有效的应用间相关度计算方法，同时缺乏有效的手段能够自动获取用户的兴趣爱好。针对以上问题，本项目研究了面向手机应用的个性化图推荐模型构建问题，该模型以应用间的相关度计算结果为基础，同时通过结合用户间的相互关注能够挖掘出用户隐含的兴趣点进而向用户推荐更加适合的应用。借助腾讯公司提供的应用宝数据，本项目还能够利用用户的行为数据（包括用户的点击、下载、评价行为）进一步提升推荐模型的准确度。结合相应的研究成果，本项目共发表论文11篇，申请专利2项，发表专著1项，获得人工智能学会创新创业奖1项，培养博士生4和硕士生12人。项目的研究成果可以广泛应用于包括企业、媒体的社交网站中用于挖掘好友之间的关系，并借助相互关系提升推荐的精准度，提高用户的满意度，同时推荐的结果能够随着用户兴趣的变化而动态改变，极大的满足用户和企业需求。