The Collaborative Search by Tag-Based User Profile in Social Media

Haoran Xie, Xiaodong Li, Jiantao Wang, Qing Li, Yi CaiThe Scientific World JOURNAL 2014 · DOI

TL;DR

This paper introduces a "Collaborative Search" approach for social media, leveraging the idea that users with similar interests can help each other find relevant information. By analyzing user and resource profiles based on tags, the method effectively groups similar users and uses this community information to improve search result ranking, addressing the challenge of information overload in the vast landscape of social media data.

What Problem Does This Paper Address?

The explosion of user-generated content on social media platforms like Delicious, Flickr, and YouTube has led to a significant problem of "information overload." Users are faced with a tremendous volume of data, making it challenging to discover the specific information they need. Traditional web search methods are often insufficient for this dynamic environment. This paper tackles this by proposing a collaborative search approach that utilizes the collective intelligence of users. The core idea is that users with shared interests, identified through their tagging behavior, can collectively improve search relevance for each other. The authors aim to develop a novel approach to profile users and resources, measure user similarity, and ultimately enhance search performance in social media.

Methods

The proposed "Collaborative Search" method is built upon three main subprocesses:

  • User and Resource Profiling: The paper adopts the "bag-of-tags" (BOT) paradigm. User profiles are represented as vectors of tag-value pairs, where the value signifies the user's degree of interest in a particular tag. Similarly, resource profiles are vectors of tag-value pairs, with the value indicating the tag's relevance to the resource. Various weighting schemes (TF, TF-IRF, BM 25, NTF) are considered for these values.
  • User Community Discovery: To efficiently group similar users, the authors propose a lightweight, two-stage approach:
  • Off-Line Clustering: This stage involves pre-clustering users into communities. They explore different similarity metrics (Jaccard, Ochiai for tag-level; Euclidean, Manhattan for value-level) and propose a "hybrid-level distance" that combines both tag and value information for more robust similarity measurement. K-means clustering is used to form these communities.
  • On-Line Discovering: This stage dynamically adapts to the user's current search context. It first checks if the current query is relevant to the user's existing profile. If not, it discovers a new, more relevant community for the user. The user's profile is then updated with the query terms. This approach aims to be more scalable than purely on-line community discovery methods.
  • Collaborative Ranking: Once users, queries, communities, and resources are defined, a ranking function (using cosine similarity) is employed to score resources. This function considers the user, their community, and the resource to produce a final ranking, aiming to surface the most relevant results.

Datasets

The paper evaluates its proposed method using two real-life datasets:

  • FMRS: This dataset contains data related to "Cooking recipes."
  • Movielens: This dataset comprises data related to "Movies."

These datasets were chosen for their different domains and scales, allowing the authors to test the method's performance across various applications. The data was split into 80% for training and 20% for testing.

Key Insights and Ideas

  • The Power of Hybrid Similarity: The research strongly suggests that combining both tag-level (what tags are used) and value-level (how relevant those tags are) information leads to a more accurate measure of user similarity than relying on either alone. This is a crucial insight for understanding user preferences in social media.
  • Scalability vs. Accuracy Trade-off: While on-line community discovery methods can achieve slightly better accuracy, the proposed off-line clustering followed by on-line adaptation offers a significant improvement in scalability and time complexity, making it more practical for large-scale social media applications.
  • NTF Paradigm and Hybrid Distance are Optimal: The experiments highlight that the Normalized Tag-Frequency (NTF) paradigm for profile construction and the hybrid-level distance metric for user similarity measurement yield the best results.
  • Community Information is Key: The results consistently show that incorporating community-based information significantly improves search performance compared to methods that only consider individual user profiles.

Key Results

  • P@N Performance: The "Collaborative" method (the proposed approach) consistently performed well, achieving the second-best P@N (Precision@N) scores across both FMRS and Movielens datasets, closely trailing the "Community-aware" baseline. The gap was between 0.7% and 2.3%.
  • MRR Performance: The Mean Reciprocal Rank (MRR) results mirrored the P@N trends, indicating that the Collaborative method is effective in quickly surfacing relevant resources.
  • Baseline Comparisons: The "Social" baseline (tag-level only) performed worse than "Collaborative," reinforcing the importance of value-level information. The "Profile-based" method, which ignored community information, performed the worst.
  • Optimal Settings: The NTF paradigm and hybrid-level distance were identified as the most effective settings for the Collaborative method, achieving the highest MRR values.

Limitations and Future Work

  • Limitations: The authors acknowledge that their "Collaborative" method's performance is slightly less accurate than the "Community-aware" baseline, which uses on-line clustering. However, they emphasize the significant scalability advantage of their approach.
  • Future Work: The authors plan to investigate the most important features of their ranking score function ($\theta$) to further enhance search quality in social media.

Summary updated

Abstract

Recently, we have witnessed the popularity and proliferation of social media applications (e.g., Delicious, Flickr, and YouTube) in the web 2.0 era. The rapid growth of user-generated data results in the problem of information overload to users. Facing such a tremendous volume of data, it is a big challenge to assist the users to find their desired data. To attack this critical problem, we propose the collaborative search approach in this paper. The core idea is that similar users may have common interests so as to help users to find their demanded data. Similar research has been conducted on the user log analysis in web search. However, the rapid growth and change of user-generated data in social media require us to discover a brand-new approach to address the unsolved issues (e.g., how to profile users, how to measure the similar users, and how to depict user-generated resources) rather than adopting existing method from web search. Therefore, we investigate various metrics to identify the similar users (user community). Moreover, we conduct the experiment on two real-life data sets by comparing the Collaborative method with the latest baselines. The empirical results show the effectiveness of the proposed approach and validate our observations.
BibTeX
@article{xie2014collaborative,
  author = {Haoran Xie and Xiaodong Li and Jiantao Wang and Qing Li and Yi Cai},
  title = {{The Collaborative Search by Tag-Based User Profile in Social Media}},
  journal = {The Scientific World JOURNAL},
  year = {2014},
  doi = {10.1155/2014/608326},
}
Figure