Different Degrees of Explicitness in Intentional Artifacts: An Exploratory Study of User Goals in a Search Query Log

Markus Strohmaier, Peter Prettenhofer, Mathias LuxCSKGOI 2008 · OpenReview

Abstract

Traditional search query classification focuses on categories like "informational" or "transactional," often overlooking the clarity with which a user expresses their underlying goal. This study introduces the concept of "intentional explicitness"—the degree to which a query explicitly states a desired outcome. By analyzing 20 million queries from the AOL search log using automated linguistic pipelines, we demonstrate that explicit queries represent a small but distinct subset of search behavior. Our findings reveal that users with explicit goals gravitate toward specific "goal-centric" web destinations and that explicitness is an orthogonal dimension to existing search taxonomies.

Introduction

Search engines are frequently described as "databases of intentions," yet they often struggle to bridge the gap between keyword strings and actual human goals. A query like "car" is implicitly intentional; the user's specific objective remains hidden. Conversely, "buy a used car in Miami" is explicitly intentional, providing a clear target for the search engine to satisfy.

The authors argue that understanding this spectrum of explicitness is vital for improving search relevance. If a user is vague, the engine must rely on probabilistic guesses; if the user is explicit, the engine has a "gold standard" to meet. However, identifying these goals at scale is difficult due to the "keyword salad" nature of search logs and the linguistic artificiality of query data.

Examples of queries correctly classified as having explicit intentionality.

Methodology

To identify explicit queries within the massive 2006 AOL search log (20 million queries from 650,000 users), the researchers developed an automated classification pipeline:

  1. Linguistic Processing: Queries were processed using Part-of-Speech (POS) tagging. A bigram tagger trained on the Penn Treebank was used, supplemented by a suffix tagger to handle the non-standard vocabulary typical of search logs.
  2. Supervised Classification: A Naive Bayes classifier was trained on POS trigrams (e.g., Verb-Determiner-Noun). The objective was to create a "condensed dataset" with a high density of explicit queries rather than achieving perfect recall across the entire log.
  3. Feature Analysis: Chi-squared analysis identified predictive patterns, such as queries beginning with "Wh-adverbs" (how, where) or specific verb-noun structures.

Key Insights and Results

The Rarity of Clarity

Explicit queries are surprisingly uncommon. The study estimates that only 1.69% to 3.01% of all queries in the AOL log are truly explicit. This suggests that most users have internalized a "bag-of-words" search style, even when they have specific goals in mind.

Goal-Centric Destinations

One of the most significant findings is that users issuing explicit queries interact with a different segment of the web. While the general population primarily clicks on major portals like Google, MySpace, or Yahoo, the "explicit" group shows a marked preference for "goal-centric" sites.

Distribution of explicit vs. implicit intentional queries across top web domains.

As shown in the data, sites like ehow.com (instructional guides), 43things.com (goal sharing), and medhelp.org (medical advice) receive a disproportionately high volume of explicit queries compared to general-purpose sites. This confirms that explicitness is a unique dimension of search behavior that correlates with specific user needs.

Linguistic Patterns

The most frequent verbs driving explicit queries include make, get, buy, find, live, play, use, and write. These verbs signal active intent. However, the researchers also identified a "cultural artifact" trap: queries like "recovering the satellites" may appear to be explicit goals but are actually song titles or movie names, necessitating domain-specific filtering to avoid false positives.

Discussion and Limitations

The study acknowledges that treating explicitness as a binary (explicit vs. implicit) is a simplification of what is likely a continuous spectrum. Furthermore, POS tagging on short queries (1-2 words) is inherently noisy, leading to the exclusion of a large portion of the dataset from deep analysis.

Future research should focus on:

  • Fine-grained Explicitness: Moving beyond binary classification to a multi-level scale of intent clarity.
  • Interactive Refinement: Developing search interfaces that help users transition from implicit to explicit queries through guided "goal refinement" dialogues.
  • Domain Knowledge Integration: Using external APIs (music, movies, products) to better distinguish between literal goals and cultural references.

Conclusion

Intentional explicitness is a unique and measurable dimension of search behavior. By identifying the small fraction of queries where users clearly state their goals, search engines can better understand the "gold standard" of user satisfaction and tailor results toward the goal-centric corners of the web that these users naturally seek.

Summary updated

BibTeX
@inproceedings{strohmaier2008different,
  author = {Markus Strohmaier and Peter Prettenhofer and Mathias Lux},
  title = {{Different Degrees of Explicitness in Intentional Artifacts: An Exploratory Study of User Goals in a Search Query Log}},
  booktitle = {CSKGOI},
  year = {2008},
  url = {https://openreview.net/forum?id=wSuByJ0VMd},
}
Figure