Citation Context

...ractions. One of earlier works on leveraging user interaction to infer document relevance is by Fox et al. [22], who showed that the user behavior such as dwell time, has strong association with explicit judgments of satisfaction and can be served as implicit feedback to improve the search experience. Following this research, Xu et al. [48], proposed a method to estimate personalized word-level dwell time from the user’s search and browse history. The proposed model can be used to estimate personalized dwell time on new documents to customize search result ranking. More recently, White et al. [46] proposed to model cohorts of users who conduct similar tasks to improve the search personalization. Recommender systems. The proactive system also shares many properties of typical recommender systems [36]. Similar to such systems, we aim to present the most relevant content to the user in the absence of an explicit query. However, the recommendations in the proactive system are for a smaller set of items that are highly repeatable and heterogeneous, and need to be personalized and contextualized in its ranking. Furthermore, we do not compare preferences across users and items as is common in...

Citation Context

...also used in different IR applications to estimate the probability of translating from one sequence of words to another sequence in the same language (e.g. (Gao et al., 2012), (Gao et al., 2010) and (=-=White et al., 2013-=-)). More formally, the similarity between two sequences of words, Q = {q1, ..., qI} and S = {s1, ..., sJ}, can be defined as: P (S|Q) = I∏ i=1 J∑ j=1 P (si|qj)P (qj |Q) (2) where P (q|Q) is the unigra...

Citation Context

...ch and Retrieval Keywords:Search Personalisation; Latent Dirichlet Allocation; Query Log; Re-ranking 1. INTRODUCTION AND BACKGROUND Recently, search personalisation has attracted increasing attention =-=[1, 3, 5, 8, 9]-=-. Unlike classical search methods, personalised search systems use personal data about a user to tailor search results to the specific user. This information can be considered as a user profile. A wid...

Citation Context

... have also shown some limited utility in retrieval settings.sGroupization has shown promise in laboratory settings [30], tasksmodels to find those engaged in similar tasks have yielded strongsresults =-=[37]-=-, and there have even been attempts to use CF more directly in search result ranking [25]. However, there has been no detailed study of applying cohort models to enhance Web-scale searchspersonalizati...

Citation Context

...t Topics, Search Personalisation, Re-ranking. 1 Introduction As one of the key components in advanced search engines (e.g., Google and Bing), Search Personalisation has attracted increasing attention =-=[1,9,12,15,16,19]-=-. The personalisation is expected to improve the usefulness of search algorithms. Unlike the search methods which don’t use personalisation, personalised search engines utilise the personal data of ea...

Citation Context

... by a wide variety of past personalization research that utilize behavioral signals. Search engine researchers have utilized user profiles to provide personalized search results [9], [11], [4], [30], =-=[35]-=-. Usually user profiles are represented as vectors in a high dimensional space [1], [23], with vectors denoting users’ preferences on different items, (e.g., web pages, movies, or social media posts.)...

Citation Context

...ation need actually spans multiple sessions, or even multiple days [6, 3]. Consequently, researchers have urged a new definition of search tasks to overcome the drawbacks of the previous session definition, hoping to better understand user search interests. Our previous research on search task identification has demonstrated the power of task, which can capture user information need more precisely than session [4, 6, 5]. It has also shown to be very effective on improving search relevance [2], as well as the applications on query suggestions [4], user behavior analysis [3] and personalization [7]. The objective of this demo is to unify our previous effort on task identification and user interests modeling, by providing several web services that takes user queries as input and output (1) identified user search tasks with semantic information, and (2) user topical interests. By doing so, we hope to decipher user search genome, provide the community with an open platform of resources with models learnt from large corpus of commercial search engines, as well as to trigger a new horizon of research effort towards this direction. 2. DESCRIPTION OF OUR SERVICES AND DEMO We provide three serv...

Citation Context

...d clicks as an indicator of relevance and considered MRR (Mean Reciprocal Rank) of these clicks as the principal performance measure. The same labelling method and performance measures were chosen in =-=[22]-=-. In this paper White et al. improved ranking of results for the current user query by using behavioral data of the other users in the same cohort. In this study we consider the same evaluation method...

Citation Context

...se previous studies [Shen et al. 2005; Tan et al. 2006] found that users’ query history can also help improve search performance and SERPs are often used as an expanded representation of the queries [=-=White et al. 2013-=-]. There is another possible reason that G+HMTI performs worse than G+HMTI-SERP. When logging theMTIs, our system can detect the corresponding HTML element. However, it cannot recognize which part of ...

Citation Context

...d on the models of her previous or current interests. Teevan et al. [7] find that retrieval performance can be improved as more data becomes available about the searcher’s interests. And White et al. =-=[11]-=- investigates the effectiveness of a task-based approach in predicting the searcher’s interests. They explore the value of modeling current task behavior, finding a significant opportunity in leveragi...