Quantum Interaction Models For Exploratory Information Retrieval

Information retrieval has widespread applications,such as search engine,recommenders and dialogure system,etc.Recently,the multi-round,interactive and exploratory information retrieval mode has become a mainstream that people retrieve and access information through search engine.However,our understanding for this new information retrieval mode is still not enough and the study about it is also relatively weak.In order to address these problems,we conduct a series of researches as follows:Firstly,we will study the mechanism of users' relevance judgment under the exploratory information retrieval mode in order to gain in-depth understanding of this new IR mode.To this end,in this thesis,we carried out two aspects of research: first,we propose to understand a multidimensional relevance model by analyzing Bing query logs quantitatively;Second,we carried out a series of user study in order to identify and verify if there exist non-classic(quantum-like)cognition phenomena in the process of users' relevance judgment.Secondly,we propose two exploratory IR models based on subspace theory.First,we propose a user profiling method based on subspace which can capture users' multiple interests in a unified framework;Second,we further present a model to capture the multidimensional relevance for exploratory IR with subspace.In this model,each decision dimension is expressed as a standard basis vector,and seven decision dimensions jointly form a subspace.Thirdly,we propose a exploratory session search model based on quantum interference.Inspired by the quantum interference theory,in this thesis,we proposed a quantum query expansion model for exploratory information retrieval.In this model,the user's information needs can be simultaneously influenced by explicit information needs(current query terms)and implicit information needs(historical query terms).Compared with classic fusion models,our models can capture the interference relationship between explicit and implicit information.Finally,we propose to extend a quantum language model for exploratory session search.In the information exploration process,user's information need is influenced by the context information,and shows obvious dynamic evolution characteristics.In order to capture the dynamic information need,in this thesis,we propose to extend the quantum language model for a typical exploratory IR scenario,i.e.,Session Search.Compared with classic approaches,our models can capture the semantic features of dynamic information need.Overall,the study in this thesis has significant scientific meaning,in terms of deepening our understanding on users' relevance judgment mechanism and bringing new idea to the field of information retrieval.