Font Size: a A A

The Research Of Model About Group-User Social Strength Evaluation And Algorithm About Location-Recommendation

Posted on:2017-05-06Degree:MasterType:Thesis
Country:ChinaCandidate:F XiaoFull Text:PDF
GTID:2348330488996347Subject:Software engineering
Abstract/Summary:PDF Full Text Request
With the rapid development of Internet, location-based service is constantly changing people's life and becoming more intelligent, more personalized and more socialized. One of the most typical examples is the Location-based Social Networks(LBSNs). LBSNs added real-time information of location on the basic of the existing social networks, joining network and reality together, and more truly reflecting users' preferences and habits. A large amount of spatio-temporal data is created including trajectory data, media data with geographical tags and check-in data. Through the analysis of massive spatio-temporal check-in data, the deep useful information is found, and the users can be fully understood, bring an individuation and unique advises as recommendations to the users.The user social strength evaluation and location-recommendation are important research directions in the research field of LBSNs. Given group-user has a more sustained and stable interaction in the social network. The check-in data is used to research group-user social strength evaluation and popular location-recommendation in paper. Group-user social strength evaluation is a measure of users' social closeness in the social network, and location-recommendation refers to a suggested position where user is likely to go but haven't been.This article proposes two groups-user social strength evaluation models so as to more effectively measure the social closeness of group-user: Shannon Entropy-based Model(SEM) and Renyi Entropy-based Min-max Model(REMM). In order to decrease the impact of frequent coincidence, Shannon entropy is used in SEM to measure the diversity of co-occurrence, and weighted frequency is used on location entropy to increase the impact of co–occurrences at less crowded places. To solve the problem of data sparseness and reduce the impact of coincidence, on the basis of considering co-occurrences position characteristics, REMM adjusts the co-occurrence's influence by changing the q value, and ultimately gets the group-user social strength.For the purpose of more effectively recommending popular location for users, two position-recommendation algorithms are proposed in the paper: Location-Recommendation based on Two-user Relationship(LRTR) and Location-Recommendation based on Group-user Relationship(LRGR). Through analyzing the history of the user's check-in data, using user-based model, establishing the relationship of geographical location between the users, LRTR quantifies the user's check-in behavior, calculating user similarity between users, predicting user preference of every point in accordance with the similarity of user's preferences, and recommending popular sites according to the recommended sequence. By analyzing group-user check-in habits, calculating groups value, computing group affect and user's preference of check-in location, and the higher score position of other members in the group which has the highest impact have visited is recommended for users in LRGR.Using check-in data sets from Brightkite and Gowalla social networking websites, the paper evaluates the efficiency of the proposed model and algorithms with experiments under a wide range of parameter settings, verifying the effectiveness of the algorithms. Experiment results show that, REMM's accuracy is better than SEM in evaluating group-user social strength, LRGR's accuracy and stability are superior to LRTR in location-recommendation. Meanwhile, REMM and LRGR are able to adapt to one million data sets, and show the good scalability under different scale data sets.
Keywords/Search Tags:social networks, spatio-temporal check-in data, group-user, social strength, location-recommendation
PDF Full Text Request
Related items