Influence Analysis in the Blogosphere

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.

💡 Research Summary

The paper tackles the problem of measuring and exploiting user influence within the blogosphere, focusing on a large‑scale Japanese blogging platform called BIGLOBE. The authors argue that previous work relied on indirect, heuristic signals such as hyperlink structures, comment counts, or follower numbers, which are noisy and often fail to capture the true causal effect of one user on another. To overcome this limitation, they exploit raw web‑server logs that record every HTTP request made by registered members. By carefully parsing these logs, they are able to reconstruct two fundamental actions: (1) writing a blog post (detected via POST requests to specific URL patterns) and (2) reading another member’s post (identified through GET requests, session cookies, and referrer information). A session‑segmentation algorithm groups consecutive page views into a single “reading session” and estimates the user’s attention span based on dwell time, thereby filtering out bots and accidental clicks.

The core of the influence‑detection framework consists of a two‑step procedure. First, a temporal precedence rule flags a “trigger‑response” pair when user A reads a post authored by user B and, within a predefined time window (24 hours in the experiments), user B publishes a new post. Second, the authors assess whether the observed frequency of such pairs exceeds what would be expected by chance. They generate a null distribution through bootstrap resampling of the entire event stream, compute p‑values for each pair, and retain only those with p < 0.05 after applying a False Discovery Rate (FDR) correction. This statistical rigor dramatically reduces false positives compared with naïve co‑occurrence methods.

Having built a directed influence graph, the study proceeds with two complementary analyses. For topic‑level insight, the authors run Latent Dirichlet Allocation (LDA) on all posts, extracting 30 topics. They then construct a separate influence sub‑graph for each topic and examine structural properties. In technology‑related topics, influence is highly concentrated: the top 5 % of bloggers generate more than 60 % of the influence edges, mirroring traditional “power‑law” expectations. In contrast, lifestyle and travel topics display a more egalitarian pattern, with many mid‑size bloggers exerting comparable influence. Member‑level analysis compares the log‑derived influence scores with conventional metrics such as post count, follower count, and account age. The correlation is modest (≈ 0.45), indicating that many users who appear marginal by traditional standards actually wield significant influence in specific contexts—a finding that uncovers “latent influencers.”

To demonstrate practical value, the authors integrate the influence graph into a recommendation engine. The baseline is a content‑based collaborative‑filtering system that suggests posts based on textual similarity. The enhanced model augments similarity scores with an “influencer weight” derived from the influence graph. Offline experiments on a held‑out test set show that the hybrid approach improves click‑through rate (CTR) by 12 % and average session duration by 9 % relative to the baseline, with especially strong gains for cold‑start users who have little interaction history. This confirms that influence information provides a robust, complementary signal for personalization.

In summary, the paper makes several notable contributions: (1) a reproducible pipeline for extracting high‑fidelity user actions from raw server logs; (2) a statistically validated method for inferring directed influence relationships; (3) empirical evidence that influence is topic‑dependent and often diverges from traditional popularity metrics; and (4) a concrete demonstration that influence‑aware recommendations outperform standard approaches. The authors argue that their methodology is not limited to blogging platforms; any online service that records detailed access logs (forums, e‑commerce sites, news portals) could adopt the same framework to uncover hidden influence dynamics. Future work is suggested in the direction of real‑time influence detection, multimodal content integration (images, video), and exploring causal inference techniques beyond temporal precedence to further refine the understanding of how digital opinions propagate.