Search engines such as Google assist users by predicting what they are looking for when they start typing a query. This type of assistance comes in the form of query auto-completion, e.g., when a user types the letters ‘face’, a search engine may suggest the completion ‘facebook’. But how can search engines predict what users are really looking for?
Fei Cai, PhD student at ILPS (Informatics Institute at the Faculty of Science), developed a new method for query auto-completion, which beats the current state-of-the-art by up to 7%. Cai does this by considering temporal aspects and users’ personal search history. The results of his research were recently presented at the 23rd edition of the international conference ‘ACM Conference on Information and Knowledge Management’ in Shanghai.
Today’s query auto-completion suggestions are primarily based on frequencies: when people more often search for ‘facebook’ than ‘face off’, search engines are more likely to suggest ‘facebook’ once the user types ‘face’. Cai notes that with this frequency-based approach two important aspects are overlooked: the temporal aspect and the user’s personal profile.
The period in which a query is issued plays an important role in query auto-completions, as some queries show seasonal patterns or periodicity. The query ‘halloween’, for example, will be more prevalent in late October than the rest of the year, so when a user types ‘hallo’ in that period, it is likely that the appropriate completion is ‘halloween’.
Additionally, the personal profile of the user is relevant to query auto-completion. For example, the user’s location can be important in deciding whether ‘University of A’ should be completed to ‘University of Amsterdam’ or to ‘University of Antwerp’. Cai’s new method takes both aspects into account: query periodicity, and the user profile, by considering previous queries of the user. His method is able to outperform the current state-of-the-art by up to 7%.
Cai’s research has been conducted at the Informatics Institute of the University of Amsterdam, under supervision of Prof. dr. Maarten de Rijke, and with the use of datasets that have been made available by AOL and the Dutch Institute of Sound & Vision. Cai’s research has been supported by the China Scholarship Council (CSC).
F. Cai, S. Liang and M. de Rijke. Time-sensitive personalized query auto-completion. In: CIKM 2014: 23rd ACM Conference on Information and Knowledge Management, November 2014, page 1599-1608.