Collaborative Filtering Meets Mobile Recommendation: A User-Centered Approach

Vincent W. Zheng, Bin Cao, Yu Zheng, Xing Xie and Qiang Yang (AAAI-10)
http://www.aaai.org/ocs/index.php/AAAI/AAAI10/paper/viewFile/1615/1964
The paper “Collaborative Filtering Meets Mobile Recommendation: A User-Centered Approach” introduces an approach, user-centered collaborative location and activity filtering (UCLAF), to extract data from many users and apply collaborative filtering to find like-minded users and like-patterned activities at different locations. These findings from social information can help to do mobile recommendation on location and activities. Different from previous work, Vincent et al. model the user-location activity relations with a tensor representation; and suggest that a regularized tensor and matrix decomposition are better to address the sparse data problem. The result of this paper shows that UCLAF has a better performance compared to some state-of-the-art solutions.
With the rapidly increase of internet data about location, location tracking services such as GPS are very good resources which can be exploited to make timely and targeted recommendations for users on places which users might be interested, and activities they are likely to enjoy at a specific place. The authors aim to mine users’ GPS trajectories based on their partial location and activity annotations, and then build the model to recommend locations and activities to individual users. Because of considering that activities are location-dependent, it motivated them to build their model that help to make inferences collaboratively rather than separately.
Vincent et al. first extract users’ location and activities annotations and then model them as a user-location-activity tensor which presents the information that users visited some place and did something there. For recommendations, they use collaborative filtering method to fill the tensor. Regularized tensor and matrix decomposition are applied for collaborative filtering solution. The tensor is decomposed into some low dimensional matrices with respect to tensor entity (i.e., users, locations and activities):
  • User-user matrix: encodes the user-user similarities in a social network
  • Location-feature matrix: each feature refers to the number of point of interests
  • Activity-activity matrix: represent the activity-activity correlations
  • User-location matrix: extracted from GPS data to model the user-location visiting relations in the case when a user visited some place but we don’t have information about what she was doing there.
After these low-dimensional representations are obtained, they reconstruct the tensor by filling all the missing entries in the tensor so that they have enough information to do recommendations for all users.
For experiments, the authors collected data from 164 users who carried GPS devices to record their trajectories from April 2007 5o Oct. 2009 in Beijing, China. The dataset consists of 12,765 GPS trajectories with a total length over 139,310 kilometers. They consider 5 different types of activities including “Food& Drink”, “Shopping”, “Movies & Shows”, “Sports & Exercise” and “Tourism & Amusement”. After clustering, they obtained 168 locations, 164 users, 5 activities and 14 location features.
To evaluate their system, Vincent et al. employ 5 baselines for comparison: user-based CF (UCF), location-based CF (LCF), activity-based CF (ACF), unifying user-location-activity CF (ULA), high-order singular value decomposition (HOSVD). They used two metrics:
  • RMSE (root mean square error): measure the tensor reconstruction loss on the hold-out test data
  • nDCG (normalized discounted cumulative gain): measure the ranking results of our retrieved location/activity list
The result from evaluating the system on a GPS dataset shows 19% improvement on location recommendation, and 22% improvement on activity recommendation over the simple memory-based CF baseline (i.e. UCF, LCF, ACF).
In the scope of this paper, we can see that leveraging social information access helps to improve recommender systems’ performances; especially, in the cases the data of users is sparse. Collaborative filtering techniques are widely applied by gathering data from many users and online resources to find like-minded users and like-patterned activities. The proposed approach of Vincent et al. demonstrated the usefulness of social information. In particular, the paper shows how to answer two common questions in location-activity recommender systems: “If I want to do something, where should I go?” and “If I will visit some place, what can I do there?”.

Comments

Post a Comment

Popular posts from this blog

Influences on Query Reformalution in Collaborative Web Search

Zhen Yue, Shuguang Han, Daqing He, and Jiepu Jiang ( Computer-2014 ) http://ieeexplore.ieee.org/abstract/document/6766167/ In the paper “Influences on Query Reformation in Collaborative Web Search”, Yue et al. investigate a study to analyze the influences of collaboration between members in a group for information seeking. They are motivated to study (1) the influence of collaboration on search query formulation and reformulation; (2) similarities and differences exist in terms of query reformulation for collaborative searches compared with individual searches; and (3) the nature of the query task influence query reformulation in collaborative information seeking (CIS). In order to run the study, Yue et al. used CollabSearch, a Web search system developed by their group for collaborative users. The system includes three frames: a topic statement page which shows the description about the search task a group want to conduct; a web search page which displays Google search results...
Read more

Personalized Access to Scientific Publications: from Recommendation to Explanation

Dario De Nart, Felice Ferrara, and Carlo Tasso ( UMAP-2013 ) http://link.springer.com/chapter/10.1007/978-3-642-38844-6_26 In the paper “Personalized Access to Scientific Publications: from Recommendation to Explanation”, Dario De Nart et al. present a Recommender and Explanation System (RES) that is able to recommend scientific publication and show explanation to users.  In order to do recommendation, Dario De Nart et al., firstly, use the The Dikpe Keyphrase extraction algorithm introduced by Pudota (2010) to extract keyphrases from papers. Each keyphrase has a weight called keyphraseness that reveals the several lexical and statistical indicators exploited in the extraction process. Higher is the keyphraseness, more relevant is the KP. For each paper in the collection, they represent it as a conceptual graph. Each node in the graph is a term broken down from a KP. Two nodes are connected when they appear in the same KP. For instance, the KP “document retrieval” is split int...
Read more

SoRec: Social Recommendation Using Probabilistic Matrix Factorization

Hao Ma, Haixuan Yang, Michael R. Lyu, Irwin King ( CIKM-08 ) http://dl.acm.org/citation.cfm?id=1458205 In this paper, Hao Ma et al. present an approach that help to deal with very large datasets and users who have made few ratings in recommender systems. Traditional methods (e.g., Pearson Correlation Coefficient and Cosine) assume that two users have rated at least some items in common so that they can compute the similarity of these users. Thus, memory-based and model-based collaborative filtering algorithms fail to find similarities of users who have never rated any items. In daily life, people usually ask their family or friends for recommendations of new movies, songs, books, restaurants or places to visit. Therefore, the authors propose a factor analysis approach based on probabilistic matrix factorization to solve the data sparsity and poor prediction accuracy problems by employing both users’ social network information and rating records. Different with traditional persp...
Read more