Arijit De

A Fuzzy Ordered Weighted Average (OWA) Approach to Result Merging for Metasearch Using the Analytical Network Process

A metasearch engine is a Information Retrieval (IR) tool that greatly expands the scope of Internet Search. Functionally a metasearch engine passes a query to multiple search engines and combines and re-ranks results returned by them in one merged list. Result Merging is a key component of metasearch and different techniques have been applied to solve the problem. In this paper we propose a model for result merging based on Satys Analytical Network Process (ANP) and Yagers Ordered Weighted Average operator. We compare our results with four existing models for result merging, namely the Borda-Fuse, Weighted Borda-Fuse, Condorcet Fuse and the OWA model.

Weighted Fuzzy Aggregation for Metasearch: An Application of Choquet Integral

A metasearch engine is an Information Retrieval (IR) system that can query multiple search engines and aggregate ranked list of results returned by them into a single result list of documents, ranked in descending order of relevance to a query. The result aggregation problem have been largely treated as Multi-Criteria Decision Making (MCDM) problem with previous approaches applying simple MCDM techniques such as average, sum, weighted average, weighted sum. Previous research has demonstrated the effectiveness of applying Yager’s Fuzzy Ordered Weighted Average (OWA) operator and its variants in result aggregation. In this paper we propose a result aggregation model based on the Choquet Integral, called Choquet Capacity-Guided Aggregation (CCGA) model which represents an alternative way to aggregate results for metasearch using most equilibrated conditions. We compare the proposed model against existing result aggregation models such as the Borda-Fuse, Weighted Borda-Fuse, OWA and IGOWA.

An unsupervised approach to automated selection of good essays

Evaluating essays automatically has been an area of active research for some time. In this paper, we propose an unsupervised technique to select a set of good essays from a large selection of essays written on the same topic. We use a ‘bag of words’ approach which does not require deep parsing. The approach is based on the content of individual essays and the divergence of the individual essay from the collection when the collection is considered as one large essay. The approach is unsupervised and does not require any reference text to build computational learning model. We evaluate our approach on a set of essays, written by different people, on a single topic submitted to a competition internally within our organization. The approach enables selection of good essays which have a good correlation with the human based selection.

On the Role of Compensatory Operators in Fuzzy Result Merging for Metasearch

A key metasearch engine task is result merging of search results from multiple search engines in response to a user query. The problem of result merging has been widely studied as a multi-criteria decision making model (MCDM). While many MCDM techniques have been employed to create experimental models for result merging, the most notable have used fuzzy aggregation operators such as the OWA operators and its extensions and variations. In this work we study the role of applying fuzzy algebraic t-norms, s-norms and compensatory operators in fuzzy result merging for metasearch. Our results will demonstrate the superiority of compensatory operators over t-norm aggregation functions in the context of result merging for metasearch.

Fuzzy Analytical Network Models for Metasearch

Merging of search engine results is a key metasearch engine function. Most result merging models try to merge ranked lists of web documents returned by search engines in response to a user query using some linear combination approach. A few give more importance to one search engines as opposed to another based on some performance criteria. Other assign weights to documents ranks etc. However few models compare documents and search engines head to head during the process of result merging. In this paper we propose two models for result merging for metasearch, Fuzzy ANP and Weighted Fuzzy ANP that employ fuzzy linguistic quantifier guided approach to result merging using Saty’s Analytical Network Process. We compare our models to existing result merging models. Our results show significant improvements.

A rule-based Short Query Intent Identification System

Using SMS (Short Message System), cell phones can be used to query for information about various topics. In an SMS based search system, one of the key problems is to identify a domain (broad topic) associated with the user query; so that a more comprehensive search can be carried out by the domain specific search engine. In this paper we use a rule based approach, to identify the domain, called Short Query Intent Identification System (SQIIS). We construct two different rule-bases using different strategies to suit query intent identification. We evaluate the two rule-bases experimentally.

Hybrid Fuzzy result merging for metasearch using Analytic Hierarchy Process

A metasearch engines is a tool that can be used to query multiple search engines in parallel. Functionally a metasearch engine passes the user query to multiple search engines, retrieves result lists of documents from them in response and then merges the result lists into one result list for the user. A key function of metasearch engines is to merge results returned by multiple search engines. Here we propose a new model for result merging that is based on Analytic Hierarch Process (AHP) and uses the OWA operator for aggregation. Our model combines the advantages of the pair wise document and search engine analysis provide by the AHP and the fuzzy aggregation provided by the OWA aggregation operator. We compare our Hybrid Fuzzy model with the OWA based model proposed by Diaz.

Fuzzy search result aggregation using Analytical Hierarchy Process

A metasearch engines is a search engine that can be used to query multiple search engines at the same time. Typically a metasearch engine passes a user query to some other search engines, which in turn returns results in the form of ranked result lists. The metasearch engine then aggregates results returned by the search engines into a single ranked result list. Result aggregation is a well studied topic. In this paper we propose a comprehensive model for result merging t-norm Importance Guided Fuzzy Hybrid model (tIGFHM) that considers search engine prior performances during aggregation, uses Saatys Analytic Hierarchy Process (AHP) to do pair wise comparisons of document and search engines and Yagers t-norm Importance Guided OWA operator to do final result aggregation. Our experiments show that our model performs better than conventional result merging models.

Unsupervised clustering technique to harness ideas from an Ideas Portal

Supervised learning techniques have long been used to analyze unstructured natural language text documents. However, supervised learning techniques are not only computationally intensive but also often require large training corpora. Supervised techniques often fail when such training corpora is either (a) not available or (b) when available, is not statistically significant to enable learning. In many practical scenarios, unsupervised learning techniques become de-facto since the training corpus is not available. In this paper we first describe an unsupervised text analysis technique and demonstrate its usefulness in addressing a real life application to harness ideas from aggregating ideas posted on our company Ideas Portal website.

SMS Based FAQ Retrieval Using Latent Semantic Indexing

In this approach note we describe two Latent Semantic Indexing approaches to SMS based FAQ retrieval. The first approach Naive LSI is based on simply applying Latent Semantic Indexing to compute the vector distance between the SMS and Questions within the FAQ set. The second approach uses the Levenshtein distance translate the SMS language in one word token strings present within the FAQ set. Then the LSI algorithm is used to construct the LSI matrix and compute the vector distance with various questions within the FAQ database. For in domain queries the second approach outperforms the first approach with a accuracy of 26%. The first approach however does better that the second approach for cross domain with an accuracy of 8%.