[Paper Review] Content-Based Spam Filtering on Video Sharing Social Networks

Antonio da Luz, Eduardo Valle+1 more|arXiv (Cornell University)|Apr 28, 2011

Video Analysis and Summarization13 references4 citations

TL;DR

This paper proposes a content-based spam filtering approach for video sharing social networks using visual features and latent semantic analysis (LSA). It evaluates static and dynamic (motion-aware) features with and without contextual information, demonstrating that context-aware analysis significantly improves spam detection accuracy, with LSA providing additional gains, proving the feasibility of visual content-based spam filtering in real-world social networks.

ABSTRACT

In this work we are concerned with the detection of spam in video sharing social networks. Specifically, we investigate how much visual content-based analysis can aid in detecting spam in videos. This is a very challenging task, because of the high-level semantic concepts involved; of the assorted nature of social networks, preventing the use of constrained a priori information; and, what is paramount, of the context dependent nature of spam. Content filtering for social networks is an increasingly demanded task: due to their popularity, the number of abuses also tends to increase, annoying the user base and disrupting their services. We systematically evaluate several approaches for processing the visual information: using static and dynamic (motionaware) features, with and without considering the context, and with or without latent semantic analysis (LSA). Our experiments show that LSA is helpful, but taking the context into consideration is paramount. The whole scheme shows good results, showing the feasibility of the concept.

Motivation & Objective

• To address the growing challenge of spam in popular video sharing social networks that disrupt user experience and service integrity.
• To investigate the effectiveness of visual content-based analysis in detecting spam, given the lack of prior knowledge and the context-dependent nature of spam.
• To evaluate the impact of context awareness and latent semantic analysis (LSA) on spam detection performance using visual features.

Proposed method

• The method employs visual features extracted from video frames, including both static (spatial) and dynamic (motion-aware) features to capture temporal changes.
• Context is incorporated by analyzing video sequences in relation to surrounding content, such as titles, descriptions, and user interactions, to improve semantic understanding.
• Latent Semantic Analysis (LSA) is applied to reduce dimensionality and uncover hidden semantic relationships in visual feature spaces.
• The system evaluates multiple configurations: with/without LSA, with/without context, and using static vs. dynamic features.
• A supervised learning framework is used to classify videos as spam or legitimate based on the combined visual and contextual features.
• Performance is evaluated using standard metrics like precision, recall, and F1-score on a real-world video dataset.

Experimental results

Research questions

• RQ1 To what extent can visual content-based analysis detect spam in video sharing social networks without relying on prior knowledge or metadata?
• RQ2 How does incorporating contextual information (e.g., video title, description, user behavior) improve spam detection accuracy?
• RQ3 What is the contribution of latent semantic analysis (LSA) in enhancing the representation of visual features for spam detection?
• RQ4 How do static versus motion-aware visual features compare in detecting spam content?
• RQ5 Can a content-based filtering system achieve high performance in real-world, unconstrained social network environments?

Key findings

• Incorporating contextual information significantly improves spam detection performance, outperforming models that rely solely on visual features.
• The use of latent semantic analysis (LSA) leads to measurable improvements in classification accuracy by capturing semantic relationships in visual feature space.
• Motion-aware features contribute more effectively than static features in detecting spam, particularly in videos with deceptive or manipulative visual patterns.
• The combination of context and LSA produces the highest F1-score, demonstrating synergy between semantic context and feature space reduction.
• The overall system achieves strong performance, confirming the feasibility of content-based spam filtering in real-world video sharing platforms.