Distinguishing humans from robots in web search logs: preliminary results using query rates and intervals

Reviewer: Donald Harris Kraft

This paper raises an interesting issue about distinguishing between humans and robots performing a search, when processing Web search logs. This may not seem like an important issue, but it is necessary to ensure proper evaluation of these logs. Moreover, one can also use the methodology employed in this paper to assess the impact of robot searches on the search engines. Duskin and Feitelson note that past work generally considered that humans do not employ Boolean operators as extensively as software agents, such as those involved with meta-search engines. In their analysis, the authors use a variety of measures, including the number of queries submitted (average rate), the minimal interval between successive queries, the rate at which queries are typed, the duration of sessions of continuous activity, the time of day of the queries, and the regularity of submitted queries (whether the same query was submitted very often or if the queries were submitted at regular intervals). The authors use three logs-AlltheWeb, AltaVista, and MSN-as their data sources, over a few days. They find that classification by number of queries and by minimal interval between queries is quite useful. Moreover, they test the thresholds by which classification (human or robot) is made, to find the best thresholds for the classification decision. Finally, Duskin and Feitelson note that this is only a first step. More testing is needed to find reliable classification methods, since they feel that one simple threshold is not accurate enough. Readers, especially those involved with Web search log analysis, should consider this most interesting paper. Online Computing Reviews Service