{{more citations needed|date=May 2024}} {{Recommender systems}} '''Implicit data collection''' refers to techniques in human–computer interaction and recommender systems that infer user preferences from observed behavior rather than explicit input.<ref>{{cite book |last1=Ricci |first1=Francesco |last2=Rokach |first2=Lior |last3=Shapira |first3=Bracha |editor-first1=Francesco |editor-first2=Lior |editor-first3=Bracha |editor-last1=Ricci |editor-last2=Rokach |editor-last3=Shapira |title=Recommender Systems Handbook |publisher=Springer |year=2015 | doi=10.1007/978-1-4899-7637-6 |isbn=978-1-4899-7636-9 }}</ref>
== Overview ==
Implicit data are used to construct a user model from interaction traces such as clicks, purchases, or dwell time. These signals enable information filtering and personalization in recommender systems and search.<ref>{{cite book |last=Joachims |first=Thorsten |chapter=Optimizing search engines using clickthrough data |pages=133–142 |title=Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining |year=2002 | doi=10.1145/775047.775067 |isbn=1-58113-567-X }}</ref>
In recommender systems, implicit feedback is often modeled through techniques such as matrix factorization and pairwise ranking, which treat user interactions as positive-only or preference signals.<ref>{{cite book |last1=Hu |first1=Yifan |last2=Koren |first2=Yehuda |last3=Volinsky |first3=Chris |chapter=Collaborative Filtering for Implicit Feedback Datasets |pages=263–272 |title=2008 Eighth IEEE International Conference on Data Mining |year=2008|doi=10.1109/ICDM.2008.22 |isbn=978-0-7695-3502-9 }}</ref><ref>{{cite journal |last1=Rendle |first1=Steffen |last2=Freudenthaler|first2=Christoph |last3=Gantner |first3=Zeno |last4=Schmidt-Thieme |first4=Lars|date=2012 |title=BPR: Bayesian Personalized Ranking from Implicit Feedback |journal=Proceedings of the Conference on Uncertainty in Artificial Intelligence |arxiv=1205.2618 }}</ref>
== Data sources ==
Implicit signals include behavioral and contextual data, such as:
* interaction logs (clicks, views, purchases) * dwell time and browsing patterns * contextual and device information * multimodal signals (e.g., gaze, voice, or facial expression)
These signals are typically noisy and require modeling assumptions to distinguish preference from exposure.<ref>{{cite journal |last1=Hu |first1=Yifan |last2=Koren |first2=Yehuda |last3=Volinsky |first3=Chris |title=Collaborative Filtering for Implicit Feedback Datasets |journal=Proceedings of the IEEE International Conference on Data Mining |year=2008 |bibcode=2008icdm.conf...43H }}</ref>
== References == {{Reflist}} Category:Recommender systems Category:Human–computer interaction