Exploiting User Behaviour in Prefetching WWW Documents

Abdulmotaleb El-Saddik, Carsten Griwodz, Ralf SteinmetzIDMS 1998 · DOI

TL;DR

This paper tackles the problem of increasing web traffic congestion by proposing a user-specific prefetching mechanism. Instead of blindly prefetching all linked pages, their system intelligently predicts which pages a user is likely to visit next based on the anchor text of links they've previously clicked, aiming to reduce unnecessary data transfer while maintaining similar hit rates.

What Problem Does This Paper Address?

The World Wide Web's popularity has led to significant traffic congestion, slowing down data retrieval. While prefetching and parallel retrieval mechanisms have been developed to mitigate this, they often exacerbate the problem by increasing overall traffic. This paper addresses the challenge of making prefetching more efficient by reducing the amount of "wasted" prefetched content that users never actually access. The authors argue that existing prefetching methods are too general and don't account for users' short-term interests, leading to the retrieval of unnecessary information.

Methods

The core of the proposed solution is a client-based proxy server called MOPER (MObile proxy helPER). MOPER employs a user-specific prefetching algorithm that relies on observing user behavior. The key components are:

  • Clicked-Word List: This acts as a user profile, storing keywords extracted from the anchor text of links the user has clicked. The frequency of each word is tracked, and this list is updated dynamically as the user browses.
  • Predict Module: This module compares the words in the anchor text of links on the current page with the user's clicked-word list. It then assigns a "prediction range" to URLs, indicating their perceived importance to the user.
  • Prefetch Module: This module receives the predicted URLs and their importance rankings from the Predict module. It then prefetches a limited number of the most relevant pages (e.g., up to 15) based on the user's predicted interests.

This approach differs from simpler methods by not prefetching all referenced pages or relying solely on server access statistics. It focuses on the semantic content of link anchors as a proxy for user intent.

Datasets

Not discussed in the paper. The authors mention conducting a "small survey on educational web sites related to multimedia" to gather data on link density and page sizes, but this is not presented as a formal dataset for training or evaluation.

Key Insights and Ideas

  • Anchor Text as a Proxy for User Interest: The paper's central idea is that the words used in link anchors provide a strong signal about a user's immediate interests and likely navigation path. This is a more nuanced approach than simply prefetching all links.
  • User-Specific Prefetching is Crucial: Generic prefetching strategies can be counterproductive. Tailoring prefetching to individual user behavior, even for short-term interests, can significantly improve efficiency.
  • Balancing Hit Rate and Traffic Reduction: The authors acknowledge that their approach might not achieve a higher hit rate than simpler methods but emphasizes that the reduction in induced traffic is substantial and more impactful for overall network health.
  • Static vs. Dynamic Content: The paper correctly identifies that prefetching is most effective for static web content, as dynamic content is inherently unpredictable.

Key Results

The paper presents experimental results comparing their user-specific approach with a "typical" (random 15 links) prefetching approach.

  • Hit-Rate: Both configurations achieved approximately the same hit-rate, around 60%. The user-specific approach achieved about 61.05%, while random prefetching achieved 63.86%. The authors note this difference is marginal.
  • Induced Traffic: This is where the user-specific approach shines. It reduced the overhead to 4.29 times the number of actually visited pages, whereas the typical prefetching technique resulted in 10.04 times the amount of data compared to visited pages. This represents a significant reduction in unnecessary data transfer.

Limitations and Future Work

  • Synonym Handling: The current implementation does not account for synonyms, meaning "car" and "automobile" would be treated as distinct. The authors plan to incorporate thesauri to address this.
  • Short Words: Words with fewer than four lowercase letters are ignored, which might miss relevant keywords.
  • Traffic Overhead: While reduced, prefetching still inherently generates additional traffic, which can be problematic if internet connection costs are time-based.
  • Future Work: The authors are interested in extending their proxy to act as a blocker for unwanted content (e.g., advertisements, explicit material) by defining keywords to be blocked, potentially even editing links out of presented pages. They also plan to incorporate thesauri to improve word decision-making.

Summary updated

BibTeX
@inproceedings{elsaddik1998exploiting,
  author = {Abdulmotaleb El-Saddik and Carsten Griwodz and Ralf Steinmetz},
  title = {{Exploiting User Behaviour in Prefetching WWW Documents}},
  booktitle = {IDMS},
  year = {1998},
  doi = {10.1007/bfb0055327},
}

Loading related content...

Figure