Intention based information retrieval

50 %
50 %
Information about Intention based information retrieval

Published on March 7, 2009

Author: atripathy86



This poster shows the idea behind Intention based Information Retrieval. Developed primarily by Amitava Biswas, Suneil Mohan, Jagannath Panigrahy, Aalap Tripathy and Prof Rabi Mahapatra at the Embedded Systems Codesign Group at Texas A&M Uivversity, College Station, TX, USA

Meaning representation for Intention based search Aalap Tripathy , Amitava Biswas, Suneil Mohan, Jagannath Panigrahy & Rabi Mahapatra Embedded Systems & Co-design Lab, Texas A&M University Intention based search Meaning Representation “Precise identification of the best available information that “The fisherman wearing a green matches user’s search intention” Text shirt, caught a big trout” Representation of Meaning Concept Tree Example of a search intention 0.77 0.4 0.45 Search for “publication on role of 1858C gene variant in catch Concept Tree 0.38 diabetes in North American population” 0.5 0.3 0.4 representation fisherman big trout 0.6 0.7 A green shirt The Problem C B Computers have limited ability to interpret meaning from human generated content (e.g. text) 0.5 × A>BC + 0.5 × BC< A + 0.3 × A>C B + 0.4 × C B< A Keyword based search can not discern: Tensor + 0.3 × A>B + 0.3 × B< A + 0.1 × A>C + 0.1 × C < A “It was not the sales manager who hit the representation bottle that day, but the office worker with + 0.4 × BC + 0.1 × C B + 0.1 × A + 0.1 × B + 0.1 × C the serious drinking problem.” from Basis Vectors Terms “That day the office manager, who was drinking, hit the problem sales worker with = “fisherman >greenshirt” A> B = “fisherman > green” A> B C a bottle, but it was not serious.” (Mitchell et al., ACL, 2008) = greenshirt < fisherman” B< A = “green < fisherman” BC > A C B = “shirtgreen” = “greenshirt” BC Key Challenge How to represent and compare meaning in computers? Meaning Similarity Comparison Similarity Computation as a dot product of two tensors Search & Retrieval Process Object Key Tensor T1 = 0.38Y Z + 0.5Z Y + 0.1 X + 0.1Y + 0.1Z + ... User makes Object Collection search query “Q” Search Key Tensor T2 = 0 . 91 Y Z + 0 . 41 X Key “K” Object “O” Similarity (T1,T2) = T1 • T 2 K1 O1 K5 O5 Common Basis Vectors (CBV) between T1 and T2 : Y Z, X K7 O7 K2 O2 T1 • T 2 = 0 . 38 × 0 . 91 + 0 . 1 × 0 . 41 = 0 . 38 K6 O6 K3 O3 K9 O9 Search Key “Q” To pair up matching coefficients of T1 & T2, we use Bloom Filters K8 O8 K4 O4 Similarity Computation using Bloom Filters (BF) Steps for searching 1. Tensor is encoded in a Bloom Filter and a Coefficient Table 2. Object Key BF (T1) & Search Key BF (T2) are BITwise ANDed 1. The meaning is represented in the form of a concept tree 3. Coefficients are extracted from table and pairwise multiplied & summed 2. The concept tree is used as a search key 3. Intention (search) key is compared with all object keys Basis Vector X BF of BF of BF of 4. Objects with matching keys are retrieved Object Key (T1) Coefficient Table CBV T2 T1 F1(X)= 1 Hash(X) Coefficients BF Indices 1 1 1 B12A34 0.38 {1,3,7,…} 0 0 0 F2E123 0.5 {3,8,10,...} F2(X)= 3 Experimental Results 0 1 0 E2E424 0.1 {1,5,6,…} 1 0 0 56EF78 0.1 {6,7,8,…} Tensor model represents meaning better than : BITwise 1 1 1 .. … … TF-IDF model in 95% cases & AND 1 1 1 ontology based vector model in 92% cases. Fj(X)= k 1 1 1 Search Key (T2) Coefficient Table Contributions 0 1 0 Hash(X) Coefficients BF Indices Technique to represent meaning in computers 0 1 0 B12A34 0.91 {1,3,7,…} 1 1 1 E2E424 0.41 {1,5,6,…} Comparison of meaning now possible

Add a comment

Related presentations

Presentación que realice en el Evento Nacional de Gobierno Abierto, realizado los ...

In this presentation we will describe our experience developing with a highly dyna...

Presentation to the LITA Forum 7th November 2014 Albuquerque, NM

Un recorrido por los cambios que nos generará el wearabletech en el futuro

Um paralelo entre as novidades & mercado em Wearable Computing e Tecnologias Assis...

Microsoft finally joins the smartwatch and fitness tracker game by introducing the...

Related pages

Web mining based on VIPS in intention-based information ...

This paper introduces a VIPS (Vision-based Page Segmentation) based Web mining method which aims to user intents based retrieval. It firstly grasps ...
Read more

Semantic Indexing for Intention-based Retrieval (PDF ...

Official Full-Text Publication: Semantic Indexing for Intention-based Retrieval on ResearchGate, the professional network for scientists.
Read more

User intention based personalized search: HPS(hierarchical ...

User intention based personalized search: HPS ... user's intention of search is very important information to retrieval information in aspect of ...
Read more

Information retrieval - Wikipedia, the free encyclopedia

Information retrieval (IR) is the activity of obtaining information resources relevant to an information need from a collection of information resources.
Read more

User intention modeling for interactive image retrieval

Patents Publication number ... The illustrated intention-based image retrieval system 108 includes an image ... on Research and Development in Information ...
Read more

Attention-based information retrieval. - ResearchGate

Attention-based information retrieval. Georg Buscher. DOI: 10.1145/1277741.1277987 Conference: SIGIR 2007: Proceedings of the 30th Annual International ACM ...
Read more

Attention-based information retrieval

Information retrieval evaluation based on the pooling method is inherently biased against systems that did not contribute to the pool of judged documents.
Read more

Image Retrieval based on intention with Hybrid Query Model ...

International Journal of Computer Applications (0975 – 8887) National Conference on Advance Trends in Information Technology, NCATIT- 2013 10
Read more

User Intention based Personalized Search : HPS ...

User Intention based Personalized Search : HPS ... intention of search is very important information to retrieval information in aspect of personalized Search.
Read more