relevanssista

Relevanssista is a term used in information retrieval to describe a user-centric signal in ranking systems that treats the final item a user interacts with during a session as a particularly strong indicator of relevance. By prioritizing the last-clicked or last-kept item, the approach aims to align rankings with the user's current intent, especially in short sessions where explicit relevance judgments are scarce.

Origin and usage: The word is a Swedish construction from relevans (relevance) and sista (last). It is discussed in the context of session-based ranking, click models, and learning-to-rank, where signals from the final interaction complement traditional item-level features.

Mechanism: Relevanssista can be incorporated by adjusting ranking scores with a session-level weight assigned to the last item, by treating the last interaction as a positive example in supervised learning, or by re-ranking an initial candidate set to promote the final item. The strength of the signal is controlled by a parameter to balance it against other features.

Applications: It is considered for web search, product recommendations, internal document retrieval, and Q&A systems, particularly when users have a clear, instantaneous intent at the end of a session.

Limitations and criticisms: Relying heavily on the last item can introduce bias, amplify noise, and overlook broader relevance; it risks overfitting to short or atypical sessions and can reduce result diversity if not carefully regularized.

See also: learning-to-rank, click models, session-based recommendation, user feedback signals.