[Paper Review] Influence Analysis in the Blogosphere

Michinari Momma, Yün Chi+3 more|arXiv (Cornell University)|Dec 24, 2012

Complex Network Analysis Techniques19 references3 citations

TL;DR

This paper proposes a principled framework to detect direct, high-confidence influence among bloggers by analyzing actual server log data of writing and reading behaviors in the BIGLOBE blogosphere. By leveraging temporal ordering, content overlap, and statistical testing, it identifies topic-specific and member-specific influence, demonstrating that personalized recommendations based on extracted influence significantly outperform global ones.

ABSTRACT

In this paper we analyze influence in the blogosphere. Recently, influence analysis has become an increasingly important research topic, as online communities, such as social networks and e-commerce sites, playing a more and more significant role in our daily life. However, so far few studies have succeeded in extracting influence from online communities in a satisfactory way. One of the challenges that limited previous researches is that it is difficult to capture user behaviors. Consequently, the influence among users could only be inferred in an indirect and heuristic way, which is inaccurate and noise-prone. In this study, we conduct an extensive investigation in regard to influence among bloggers at a Japanese blog web site, BIGLOBE. By processing the log files of the web servers, we are able to accurately extract the activities of BIGLOBE members in terms of writing their blog posts and reading other member's posts. Based on these activities, we propose a principled framework to detect influence among the members with high confidence level. From the extracted influence, we conduct in-depth analysis on how influence varies over different topics and how influence varies over different members. We also show the potentials of leveraging the extracted influence to make personalized recommendation in BIGLOBE. To our best knowledge, this is one of the first studies that capture and analyze influence in the blogosphere in such a large scale.

Motivation & Objective

• To address the challenge of distinguishing causal influence from mere correlation in online influence analysis.
• To develop a principled, data-driven method for identifying direct influence among bloggers using actual behavioral logs.
• To investigate whether influence varies across topics and across individual members.
• To evaluate the utility of extracted influence for personalized blogger recommendation.
• To provide a large-scale, empirically grounded analysis of influence in a real-world blog community.

Proposed method

• The framework uses server log files to extract precise sequences of blog writing and reading activities among BIGLOBE members.
• Influence is identified via a statistical test that checks whether a blogger's post follows a reading event of another blogger's post, with content overlap (keywords) as a condition.
• A topic-specific influence model is built by analyzing overlapping keywords between a reader’s post and the post they read before.
• The influence network is split into training and test sets using a hold-out method that prevents data leakage and avoids the 'bookmark effect'.
• Personalized recommendations are generated using Bayesian fusion of topic-specific influence and topic likelihood, with iOLAP and PCL-DC as key algorithms.
• Performance is evaluated using recall-at-N, comparing global (TG) and personalized (iOLAP, PCL-DC) recommendation strategies.

Experimental results

Research questions

• RQ1 Are influential bloggers different depending on the topic being discussed?
• RQ2 Does the set of influential bloggers vary from one member to another, even when discussing the same topic?
• RQ3 Can influence extracted from real behavioral logs be used effectively for personalized blogger recommendation?
• RQ4 How does content overlap and temporal ordering help distinguish causal influence from correlation?
• RQ5 What is the relative contribution of link structure versus content analysis in modeling influence?

Key findings

• Personalized influence models (iOLAP and PCL-DC) significantly outperform global recommendation (TG) in recall-at-N, confirming that influence is both topic-specific and member-specific.
• The simplified PCL model, which uses only link analysis without content, performs worse than PCL-DC, showing that content analysis is essential for accurate influence modeling.
• The statistical test framework successfully identifies influence with high confidence by combining temporal ordering and content overlap, reducing false positives from correlation.
• Topic-specific influence is empirically validated: different bloggers emerge as influential depending on the discussion topic.
• Influence varies per member: the same topic can have different influential bloggers for different users, supporting the need for personalized recommendation.
• The study demonstrates, to our knowledge, the first large-scale analysis of influence in the blogosphere using real behavioral logs with rigorous statistical validation.