一种基于效用的个性化文章推荐方法

A Personalized Paper Recommendation Method Based on Utility

近年来,个性化推荐方法逐渐成为推荐系统领域的研究热点,文章推荐就是其中的一个分支.然而目前大部分文章推荐方法还存在以下问题:首先,这些方法大多还是基于协同过滤或者基于内容的推荐,而这些方法无法满足用户因自身偏好带来的需求,它们倾向于通过获取文章的各种不同属性来代表文章,这样的获取方式存在着对文章本身理解的偏差.其次,还存在一些文章推荐方法使用搜索代替推荐,这种方法仅仅以用户的偏好为基础,并不考虑文章的内在属性和联系,反馈回的数据依赖于用户做大量的判断.这些都为文章推荐方法的研究提出了挑战与机遇.该文主要聚焦的问题是:(1)已有的文章有多大的可能性成为用户的选择.对于推荐方法来说,过去的文章更多的是通过一些文章属性来获取.如研究领域、作者姓名、文章权威性、文章发表时间、引用数量等,这些属性的数据量庞大且没有量化标准,用户不容易通过这些信息来准确获取自己想要的文章;(2)什么样的推荐方法能将用户需要的文章更好的推荐出来.由于用户的偏好是一种理性和感性的共同选择,因此推荐方法不仅需要考虑文章自身的影响力,还要考虑用户需求的程度和方向.该文通过分析用户与文章的关系,提出一种个性化文章推荐方法,该方法基于效用理论,为了得到效用函数,该文将文章属性和用户偏好作为效用函数的元素,首先分析了文章属性和用户偏好之间的关系,其次通过最优化的效用函数得到用户对文章的评价值,最后通过对不同的评价值进行比较,将相关的文章推荐给用户.这样就可以通过效用函数最终获取用户所需的文章.该文的贡献主要包括以下几方面.(1)建立了一种关于论文属性和用户偏好的组合模型,该模型是通过效用函数建立的,目标是使用户的利益达到最大;(2)效用函数中的变量来自于论文属性和用户偏好的取值,将论文属性和用户偏好同时进行量化,比以往单一的通过文章属性建立的方法更灵活,也比单一通过用户的偏好建立的方法更全面.通过实验对比利用该方法推荐的文章相对于其他方法准确率平均提高了7.17%,召回率平均提高了4.55%,说明该方法推荐的文章质量是可以保证的.

In recent years,the personalized recommendation method has gradually become a research hotspot in the area of recommendation system.Paper recommendation is one of the branches in this field.However,most of the available method of paper recommendation has some disadvantages as follow.Firstly,most of these methods are based on collaborative filtering or content based recommendation,which cannot meet the need of the different preferences based requirements of user accordingly.At mean time,these methods replace the papers by obtaining the various attributes of them.Additionally,there are some paper recommendation methods that use search instead of recommendation.These methods only were based on user’s preferences but ignore the intrinsic attributes and connections of paper,the feedback data need numerous subjective judgments.All of these problems promote the challenge and opportunity in the research of the paper recommendation.The main focuses of this paper as follow.(1)What’s the possibility of that the existing paper will become user’s choice?For paper recommendation,users are more likely to obtain papers by using keywords,such as the research filed,the author’s name,the authority of the paper,the publication time and the number of citations,etc.,these attributes of paper constitute a great volume of data but lack of quantitative criteria.It is not easy for users to obtain the papers they want through this information exactly.(2)What kind of recommendation method can be used to recommend better to meet the user’s needs?As the user’s preference is a kind of rational and emotional common choice,the recommended method needs to not only consider the influence of the paper itself,but also consider the extent and direction of user’s needs.In this paper,we have put forward a personalized recommendation method by analyzing the relationship between the user and the paper based on the utility theory,the users could get their needed papers through the optimized utility function.In order to obtain the utility function,we took the paper attribute and user preference as utility function elements,after analyzing the relationship between these two elements,we could capture the importance coefficient of paper through the optimized utility function,finally we executed a comparison of the different importance coefficient to get the appropriate recommendable articles to the user.Therefore,we can finally get the user required by the utility function.The contribution of this paper mainly includes the following aspects.(1)A combination model of paper attributes and user preferences is established,the model is constructed by utility function,and the objective is to maximize the interests of the user.(2)The variables in the utility function are derive from the attributes of the papers and the preferences of the users,the paper attributes and user preferences are quantified simultaneously,and it is more flexible and comprehensive than the method handles only one attribute in the past.As shown in the experiments result,compared to other methods,this method could improve the average accuracy rate by7.17%,as well as the average recall rate increased by4.55%,which means the quality of the recommend paper by this method could be guaranteed.