Webarc:Search and Access Strategies

Background

• The Web has become the main publication medium world-wide, covering almost every facet of human activity. However, the Web is an ephemeral medium.
• Web archives offer unique opportunities for knowledge discovery due to the richness of their contents extending over significant periods of time.
• We need effective and scalable access strategies for web archives covering significant temporal spans.

Goals

• An effective technology to explore and search for information in a web archive. The technology must enable effective information exploration and discovery.
• A framework that produces ranking and evaluation of archived web contents within a temporal context as required by the user.
• Methods to determine the relevance of a group of web objects within the temporal context. A group can consist of a series of temporally-contiguous versions of a single URL, or of web objects archived within some time span.
• A framework that allows effective search using keywords and time spans for large scale web archives.

Existing Access Methods

Problems With Existing Methods

• Inefficient handling of time-constrained search.

• Ineffective delivery of search results
  • Inadequate relevancy scoring.
    • Scoring is performed over the entire history.
  • Ungrouped search results.
    • URL is not unique in web archives – time dependent.
    • Considering different versions of the same URL tend to have similar contents, It is highly likely to have the first result page “polluted’ by multiple versions of the same URL.
    • Users can want to focus more on a specific time-period within the results.
  • Lack of a group-scoring methodology.
    • What group to show on the top is not clear without a group-scoring methodology.

Overview of Our Approach

• Efficient time-constrained search by maintaining separate inverted lists for a given time window. Click here for details.
• Scoring within a temporal context by computing term weights as a function of time. Click here for details.
• Grouping similar search results, while scoring search results as a group. Click here and here for details.

Basic Techniques

• Determine a snapshot of web contents covering a time window

SCk = { All web objects valid within a time interval [tk~tk+1) }

• Given SCs, we maintain a set of inverted lists for each SC, and build a time hierarchy or a combined multi-version tree.

Scoring within a Temporal Context

Search User Interface

Grouping Search Results

Group-wide Scoring

Grouping is good, but now which group to place first on the result page?

• Simple method : use average or highest score among members
• More effective method: compute a relevancy score as a group.
  • Instead of tf(t), we use df(t), document frequency of t in group.
  • Instead of idf(t), we use igf(t), inverse group frequency .

We extend some of the best known IR technologies for group ranking.

Publication

Song, S. and JaJa, J. Search and Access Strategies for Web Archives. in Archiving 2009. 2009: IS&T. pdf