Research On One-city Event Recommendation In Event-based Social Networks

With the rapidly development of Internet and the continuous innovation of Internet technology,social networks are becoming more and more mature and perfect.Among the many types of social networks,there is a social network that combines activities online and offline--Event-based Social Network(EBSN).Compared with the traditional social network,users can not only browse event information online,but also participate in the event offline according to the online information.With the development of time and network,the massive data produced in the Event-based Social Network makes it difficult for users to find their own interested event.Therefore,to improve the user experience,it's urgent to establish a kind of recommend system based on EBSN to make event recommendations for users.Social event recommendation is different from the traditional recommendation,which are as follows:(1)the one-off consumption characteristic of social event.Events are man-made,with a specific theme,time and location.So users can only participate once,unlike goods that be purchased repeatedly,and have no Historical evaluation records.(2)More information in EBSN that can be used to recommend.There are two kinds of social relationships in EBSN,online and offline.The former is constituted by online users of the same group,the latter by offline users of the same event.In addition,other information such as the time and location of users and events is useful and meaningful to event recommendation.According to these differences,traditional recommendation methods cannot be used directly in event recommendation.So,the focus is on event recommendation in this paper.The current social event recommendation algorithm often encounter problems such as data sparsity problem and the incomplete using of information.This paper first describes the current theories and algorithms of event recommendation and the challenges they faces.Then aiming at the aforementioned problems,a one-city event recommendation algorithm in event-based social network is given in this paper.The main word of the paper is as follows:(1)A one-city event recommendation model is given,which includes the following four parts: data procurement model,feature extraction model,learning to rank model and recommendation model.The data acquisition module is to get data from social network and to divide data into training data and candidate event.Feature extraction module is to extract five characteristics: user preferences,friends' influence,time matching degree,location matching degree and the popularity of the event theme by analyzing the training data.Learning ranking module is to convert the event recommendation problem into learning to rank problem,which can get the optimal weight W of all the features by learning to ranking all the data.Recommendation module is to recommend candidate event to users.Firstly,get the city of the user by analyzing his/her loading IP.Secondly,select his/her candidate events by matching the city of user and events.Then calculate his/her ratings of the candidate events according to the optimal weight W achieve in learning to rank model.Finally recommend top-N activities to him/her by sorting the scores of candidate events in descending order.(2)In this paper,we analyze and extract five characters: user preference,friends influence,time matching degree,position matching degree and the event theme popularity.The calculation method of each characteristic is described below.Based on the content recommendation method,user preference is to calculate the similarity between users and events.The LDA method is used to represent users and events' topic vectors,which reduces the text dimension and alleviates the data sparseness problem.To calculate friends' influence,we regard user preference as users' ratings,and use collaborative filtering method.The degree of time matching and the degree of location matching calculate the similarity between users and events in time and location,which respectively mining the behavior rules of users in time and location characteristics.Event popularity is the similarity between the candidate event and the city's popular theme.The city's popular theme is extracted in events which is the top-M events that be participated in his city recently.At the same time,the event popularity can reduce the impact of the cold start problem on event recommendation.(3)An event recommendation algorithm is given,which convert the recommendation problem into learning to rank problem.With the idea of pairwise learning,we convert training data into event sequence,including positive sequence pair and negative sequence pair,then event recommendation problem will be converted into classification problem of two classes.In order to comprehensively consider the influence of various characteristics,we improve the logical regression method,which is suitable for the pairwise learning problem.In the solving process,we choose the square loss as my loss function,adding a L2 regularization term to prevent overfitting and adding a user coefficient to reduce the influence caused by the unbalanced data.Finally,we use the batch gradient descent method to get the optimal weight W.The event recommendation method given in this paper is: using the improved ranking logistic regression method to combine the proposed five characters,such as user preference,friends influence,time matching degree,position matching degree and the event theme popularity;then calculating the scores of a user's candidate events and recommending the top-N to him/her.In order to verify the effectiveness of the method given in this paper,precision and recall was selected as the evaluation method to measure the recommendation quality.Using the event data crawled from Douban website,a lot of experiments are done to compare my method with other four kinds of methods.The results show that comparing to the single feature recommendation method,our method which combines five features has better performance;comparing with the existing four event recommendation methods,the one-city event recommendation method proposed in this paper has a better performance,effectively improving the user experience.