Spatial queries entity recognition and disambiguation

Spatial Queries Entity Recognition and DisambiguationBY: EHSAN HAMZEI

Table of contents 1- Introduction

2- Query Processing (Related Works)

3- State of the Art

4- Our approach

5- Conclusion

Introduction December 1990 >> First Search engine (W3Catalog) >> Entirely indexed by hand

September 1993 >> WebCrawler >> Finding automatically

…

January 1994 >> Yahoo!

September 1997>> Google

Introduction(Spatial Search Engine)

New Sources on the web:◦ New Search Engines for Images, Videos◦ New Search engine for geospatial data (Google Maps, Bing Maps)

February 2005 >> Google Maps

December 2010 >> Bing Maps

Query Processing (What is Query Processing?)

Search Engine two major process:◦ 1- Offline (For crawling and collection data)◦ 2- Online (Started from user’s query and end with returning the results)

Where is Query Processing?

What is Query Processing brings to us?

Query Processing and Related Works

NLP >> Natural Language Processing

ER >> Entity Recognition

Related Works:◦ Guo et al. (2009) addresses the problem of Named Entity Recognition in Query (NERQ)◦ …◦ Dalvi et al.(2014) developed a four step algorithm named Topic-specific Language Model (TLM method)

for doing Entity Recognition and Disambiguation from search queries.

Query Processing (State Of The Art)

An Example of two same query by google maps:

1- Intersection of shariati and resalat

2- Intersection of valiasr and enqelab

Proposed Approach (Definition) Spatial Query = Combination of:

◦ 1- Location Name◦ 2- Location Type ◦ 3- Spatial Relationship◦ Example : Hospitals around Resalat Square

Based On NLP (ER) We can recognize and tag these types for further processes

Proposed Approach (Algorithm) 1- Input Query > Segmentation (Top to Down)

2- Candidate ◦ 2-1 Location Name◦ 2-2 Location Type◦ 2-3 Spatial Relationship

3- Validate The Result◦ 3-1 Check that it is fully understand◦ 3-2 Check the conceptual criteria◦ 3-3 Check the logical criteria

4- Returning the result

Proposed Approach (Evaluation) Two kind of evaluations can be possible:

1- Disambiguation:◦ The average disambiguation for 100 spatial queries: 89.45%

2- According to 100 spatial queries compared to Google Maps◦ Google Maps : 72◦ Our Approach : 91

Conclusion Changing the perspective from textual to spatial

Take the spatial relationship into account◦ Make them answerable in general◦ Using them for disambiguation

Future Work:◦ Using the combination of Geocode APIs◦ Develop more sophisticated algorithm (2 or more spatial relationship)

Thanks For Your Attention