Query Suggestions Using Query-Flow Graphs

Post on 02-Jan-2016

46 views 1 download

Tags:

description

Query Suggestions Using Query-Flow Graphs. Paolo Boldi et al . WSCD 2009. Objective. Query suggestion using query-flow graphs. What is a query flow graph?. Node = query. What is a query flow graph?. Same “ search mission ”. What is a query flow graph?. Edge weight:. - PowerPoint PPT Presentation

Transcript of Query Suggestions Using Query-Flow Graphs

Query Suggestions Using Query-Flow Graphs

Paolo Boldi et al.WSCD 2009

Objective

Query suggestion using query-flow graphs

What is a query flowgraph?

Node = query

What is a query flowgraph?

Same “searchmission”

What is a query flowgraph?

Edge weight:

What is a query flowgraph?

Four transitions: Generalization Specialization Error

Correction Parallel Move

Corpus

Generate query-flow graphs: Yahoo! UK

Testing: MSN Live Spring 2006 Data Asset

Recommendation Algorithm 1

Start at the input query node Do a random walk of k iterations Recommend queries with highest Pag

eRank values

Recommendation Algorithm 2

Scale the score in Algorithm 1 by a random start

Recommend queries with highest PageRank ratio

Experiment Setup

Recommendation Algorithms: Different Query-flow subgraphs Baseline: Query-document graph by Cras

swell and Szummer Test Corpus

114 input queries in MSN Live having frequency between 700 and 1500

Query-Flow Subgraphs Queryflow-S Queryflow-SP Queryflow-SC Queryflow-SCP Queryflow-GSPC

Query-Flow Composed Subgraphs Queryflow-(S2) Queryflow-(SST) Queryflow-(SG)

Evaluation Procedure

Top 5 recommendations per query per system

5 assessors rated a recommendation as Useful, Somewhat useful, Not useful “Related to the same intent and

provides information not available in original query”

Experiment Results

Diversity Experiment

Issue top 5 recommendations per query per system to a search engine

Get top 5 URLs Compute average number of new

URLs per query

Diversity Experiment Results