Sequential Query Expansion using Concept Graph Date: 2016/4/18 - PowerPoint PPT Presentation

Sequential Query Expansion using Concept Graph Date: 2016/4/18 Author: Said Balabeshin-Kordan, Alexander Kotov Source: CIKM’16 Advisor: Jia-Ling Koh Speaker: Chih-Hsuan Tzang 1

Introduction Method Experiment Conclusion 2

Introduction Method Experiment Conclusion 3

Introduction Goal: • ConceptNet 4

Introduction Q: poach preserve wildlife Goal: 5

Introduction Query Query�Expansion�(LCE) Concept�Graphs Sequential�Concept�Expansion� two�stage-� I. Initial�Sorting�of�Concepts� related� II. Sequential�Selection�of�Concepts concepts� 6

Introduction Method Experiment Conclusion 7

Method Query Query�Expansion�(LCE) Concept�Graphs Sequential�Concept�Expansion� two�stage-� I. Initial�Sorting�of�Concepts� related� II. Sequential�Selection�of�Concepts concepts� 8

Method Latent Concept Expansion ( LCE ) : • LCE was designed to incorporate the query expansion terms from the top retrieved documents into Markov Random Fields - based retrieval models. 9

Method Latent Concept Expansion ( LCE ) : 10

Method Latent Concept Expansion ( LCE ) : • Scoring the document D with respect to query Q 11

Method Q: poach preserve wildlife Latent Concept Expansion ( LCE ) : {“poach”, “preserve”, “wildlife”} {“poach preserve”, “preserve wildlife”} {“poach preserve”, “preserve poach”, “preserve wildlife”“wildlife preserve”} 12

Method Query Query�Expansion�(LCE) Concept�Graphs Sequential�Concept�Expansion� two�stage-� I. Initial�Sorting�of�Concepts� related� II. Sequential�Selection�of�Concepts concepts� 13

Method Concept Graphs: • two way to construct concept Graphs i. use a manually created semantic network, such as ConceptNet. Only considered English concepts. 14

Method Concept Graphs: • two way to construct concept Graphs ii. use a collection itself. Only unigram concepts are used in the concept graph in this case. use Hyper - space Analogue to Language ( HAL ) similarity measure 15

Method Query Query�Expansion�(LCE) Concept�Graphs Sequential�Concept�Expansion� two�stage-� I. Initial�Sorting�of�Concepts� related� II. Sequential�Selection�of�Concepts concepts� 16

Method Sequential Concept Expansion: •Step 1: Initial Sorting Concept 17

Method Sequential Concept Expansion: •Step 2: Sequential Selection of Concepts Discard Discard Discard 18

Sequential Concept Expansion: •Step 2: Sequential Selection of Concepts 19

Method Sequential Concept Expansion: •Step 2: Sequential Selection of Concepts 20

Introduction Method Experiment Conclusion 21

Experiment statistics of the collections: 22

Experiment 23

Experiment Remove features MAP: 24

Experiment 25

Experiment 26

Experiment 27

Experiment QL - Query Likelihood retrieval model with Dirichlet prior smoothing RM - Relevance Model SDM - Sequential Dependence Model LCE - d Latent. Concept Expansion pseudo - relevance feedback ( PRF ) 28

Introduction Method Experiment Conclusion 29

Conclusion • The main contribution of this work: • A two - stage method for sequential selection of e fg ective concepts for query expansion from the concept graph. • The optimization problem of the proposed method: • Objective: having least possible number of candidate concepts. • Constraint: achieve a given precision of retrieval results . • Stages of the proposed method: • stage 1: sort the candidate concepts • stage 2: sequentially select expansion concepts 30