Research On The Service Recommendation Method With Enhanced Fairness

With the rapid development of network technology,service recommendation has entered a large scale of practical applications,especially the rise of mobile Internet,which has brought a truly expansive market for service recommendation.Early related research focuses on the basic issues such as user preference extraction and personalized service selection when service recommendation has just emerged.However,with the continuous expansion of the market scale,two new outstanding problems are faced in the practical application.The first is the fairness of service presentation.In order to win in the market competition,competitive service providers would spend lots of money on advertising.However,services provided by small and medium companies are very hard to show to the users,which leads to the unfair problem of service presentation.This problem has led to some services with potential high quality being buried and unable to be recommended,thus lead to the long tail effect.The recommendation opportunities are almost monopolized by the popular service.The so-called mass candidate services only exist in the concept of quantity but no effect in practical recommendations.Therefore,how to give all candidate services a fair chance to be selected becomes an important issue that should be considered in recommender systems.The second is the way to ensure the accuracy of recommendation based on fairness consideration.In the candidate services,popular services have a large amount of usage and rich historical feedback,which can get accurate QoS information,so that it can carry out relatively accurate evaluation of service quality.However,due to the low usage,QoS data of unpopular service are sparse or even blank,which brings difficulties to service quality evaluation.Therefore,in order to ensure the fairness of these unpopular services,accurately evaluating their service quality to ensure the recommendation effect becomes a key factor of service recommendation.Finally,the fairness coordination problem of individual evaluation and group evaluation in group recommendation.With the deepening of the marketing application of recommendation system,group recommendation becomes ever more common.In order to improve the recommendation effect,the recommendation system often divides users into tag groups explicitly or implicitly according to historical information,and users sometimes form specific groups spontaneously.When making service recommendation to a group,if the special needs of a single user cannot be met,it may lead to the failure of the group's overall recommendation.Therefore,how to coordinate the fairness between individual users,reduce the impact of individual irrational feedback,and to evaluate the effectiveness of the group objectively and accurately is another key problem in the service recommendation.Based on the above three fairness related issues,this paper aims at the corresponding recommendation strategy of popular and unpopular service,and excavates the potential good service in the unpopular service while ensuring the benefit of the popular service recommendation platform,so as to achieve a fair and accurate service recommendation.The fairness have been considered among three aspects,so as to enhance the fairness of service recommendation.This paper has achieved the following results:1)To address the fairness problem in the service selection process,a fairness-aware service selection algorithm is proposed.The algorithm performs different selection strategies for popular and unpopular services respectively.For the popular services,by mining the property of the service,the fuzzy analytic hierarchy process is used to sort the candidate service,so as to obtain the accurate selection results;For unpopular services,the fairness factor of every service is initialized based on knowable attributes,and then adjust it by a dynamic adjustment strategy.Finally,roulette algorithm is adopted for unpopular service selection.Popular services and unpopular services together constitute the results of service selection,so as to realize service selection in dynamic service ecosystem.The experimental results demonstrate that the proposed algorithm can accurately select highly related candidate services and significantly improve the fairness of unpopular services in the process of service selection.2)For the fairness problem in personalized service recommendation,a personalized service recommendation algorithm considering with accuracy and fairness is proposed.This algorithm performs different recommendation strategies for popular and unpopular services,respectively.The final recommendation list is composed of two services in a preset proportion.During the process of popular service recommendation,an improved BP neural network is used to model multi-dimensional context information,and then the bias factorization algorithm is employed to introduce the context evaluation deviation and make the rating prediction.Finally,popular service recommendation results are obtained by sorting the service prediction rating.In unpopular service recommendation,the distance between user preferences and service attributes is used to adjust fairness factors,and the problem of unpopular service recommendation is transformed into a combinatorial optimization problem by designing a fair utility function.The experimental results of restaurant service recommendation show that the algorithm can achieve better Diversity and fairness in the recommendation process.3)A fairness-aware group recommendation algorithm is presented for group users.Taking restaurant group recommendation as an example,the algorithm can improve the global satisfaction and fairness among users by service pre-filtering,aggregation strategy and ranking adjustment.First,the candidate service set is filtered through the design of service pre-filtering algorithm for group users.The candidate services that do not comply with the requirements or have a large difference in fairness are filtered out.Secondly,group relevance and group disagreement in the group consensus function are improved,by introducing the intra-group influence and the Borda counting rule.So that the recommendation results are aggregated more accurately and fairly.Finally,the fairness of recommendation result in the key context is realized by the ranking adjustment strategy based on key context.The experimental results show that the proposed method can effectively improve the global satisfaction and key context satisfaction of group recommendation,especially in terms of fairness among users.