Model adaptation via model interpolation and boosting for web search ranking

Gao, Jianfeng; Wu, Qiang; Burges, Chris; Svore, Krysta M.; Su, Yi; Khan, Nazan; Shah, Shalin; Zhou, Hongyan

doi:10.3115/1699571.1699578

Cited by 22 publications

(25 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Gao and Wu et al [Wu et al 2008;Gao et al 2009] comparatively studied the model interpolation method, i.e., linear combination of the source domain model and the target domain model, and a gradient boosting method similar to our additive model, i.e., appending trees to the source domain model to minimize the target domain prediction error only. Geng et al [Geng et al 2009] formulate the adaptation problem in web search under the framework of quadratic programming and thus a method is similar to SVM can be applied.…”

Section: Related Workmentioning

confidence: 99%

Ranking function adaptation with boosting trees

Chen

Bai

Zheng

2011

ACM Trans. Inf. Syst.

View full text Add to dashboard Cite

Machine learned ranking functions have shown successes in web search engines. With the increasing demands on developing effective ranking functions for different search domains, we have seen a big bottleneck, i.e., the problem of insufficient labeled training data, which has significantly slowed the development and deployment of machine learned ranking functions for different domains. There are two possible approaches to address this problem: (1) combining labeled training data from similar domains with the small targetdomain labeled data for training or (2) using pairwise preference data extracted from user clickthrough log for the target domain for training. In this paper, we propose a new approach called tree based ranking function adaptation ("Trada") to effectively utilize these data sources for training cross-domain ranking functions. Tree adaptation assumes that ranking functions are trained with the Stochastic Gradient Boosting Trees method − a gradient boosting method on regression trees. It takes such a ranking function from one domain and tunes its tree based structure with a small amount of training data from the target domain. The unique features include (1) it can automatically identify the part of model that needs adjustment for the new domain, (2) it can appropriately weigh training examples considering both local and global distributions. Based on a novel pairwise loss function that we developed for pairwise learning, the basic tree adaptation algorithm is also extended ("Pairwise Trada") to utilize the pairwise preference data from the target domain to further improve the effectiveness of adaptation. Experiments are performed on real datasets to show that tree adaptation can provide better-quality ranking functions for a new domain than other methods.

show abstract

Section: Related Workmentioning

confidence: 99%

Ranking function adaptation with boosting trees

Chen

Bai

Zheng

2011

ACM Trans. Inf. Syst.

View full text Add to dashboard Cite

show abstract

“…The generality of these approaches is limited either by the type of queries or in the setting (traditional TREC style) they are explored. In [13], English training data from a general domain has been used to improve the accuracy of the English queries from a Korean market. But in this particular case both in-domain and out-of-domain data are from English, hence the set of features used for the learning algorithm remain same.…”

Section: Related Workmentioning

confidence: 99%

“…Different flavors of this problem have been attempted by other researchers [10,11,12,13,14]. At a broader level, there are two possible approaches to transfer the useful information across languages depending on the availability of original queries in both the languages.…”

Section: Introductionmentioning

confidence: 99%

Fractional Similarity: Cross-Lingual Feature Selection for Search

Jagarlamudi

Bennett

2011

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. Training data as well as supplementary data such as usagebased click behavior may abound in one search market (i.e., a particular region, domain, or language) and be much scarcer in another market. Transfer methods attempt to improve performance in these resourcescarce markets by leveraging data across markets. However, differences in feature distributions across markets can change the optimal model. We introduce a method called Fractional Similarity, which uses query-based variance within a market to obtain more reliable estimates of feature deviations across markets. An empirical analysis demonstrates that using this scoring method as a feature selection criterion in cross-lingual transfer improves relevance ranking in the foreign language and compares favorably to a baseline based on KL divergence.

show abstract

“…One of the possible ways to address the above challenge is to build LSP models that learn lexico-syntactic patterns on generic and ontology rich domains and then apply these patterns on specific ontology poor domains. In line with (Gao, 2009), we respectively refer as the background domains and application domains to these two kinds of domains. Yet, in machine learning and in statistical learning data should be enough representative of the environment where learned models will be applied.…”

Section: Generic Ontology Learners On Application Domainsmentioning

confidence: 99%

“…In (Snow, Jurafsky, & Ng, 2006), all WordNet has been used as source of training examples. In this case, domain adaptation techniques must be adopted (Bacchiani, Roark, & Saraclar, 2004;Roark & Bacchiani, 2003;Chelba & Acero, 2006;Gao, 2009;Gildea, 2001).…”

Section: Adapting Semantic Network To New Domainsmentioning

confidence: 99%

Exploiting Transitivity in Probabilistic Models for Ontology Learning

Fallucchi

Semi-Automatic Ontology Development

View full text Add to dashboard Cite

Capturing word meaning is one of the challenges of natural language processing (NLP). Formal models of meaning such as ontologies are knowledge repositories used in a variety of applications. To be effectively used, these ontologies have to be large or, at least, adapted to specific domains. Our main goal is to contribute practically to the research on ontology learning models by covering different aspects of the task. We propose probabilistic models for learning ontologies that expands existing ontologies taking into accounts both corpus-extracted evidences and structure of the generated ontologies. The model exploits structural properties of target relations such as transitivity during learning. We then propose two extensions of our probabilistic models: a model for learning from a generic domain that can be exploited to extract new information in a specific domain and an incremental ontology learning system that put human validations in the learning loop. This latter provides a graphical user interface and a humancomputer interaction workflow supporting the incremental leaning loop. INTRODUCTIONGottfried Wilhelm Leibniz was convinced that human knowledge was like a "bazaar": a place full of all sorts of goods without any order or inventory. As in a "bazaar", searching a little piece of specific knowledge is a challenge that can last forever. Nowadays, we have powerful machines to process and collect data. These machines, combined with the human need of exchanging and sharing information, produced an incredibly large evolving collection of documents, partially shared with the World Wide Web. The Web is a modern worldwide scale knowledge "bazaar" full of any sort of information where searching specific information is a titanic task. Ontologies represent the Semantic Web's reply to the need of searching knowledge in the Web. These ontologies provide shared metadata vocabularies (Berners-Lee, T., Hendler, J., & Lassila, O., 2001). Data, documents, images, and information sources in general, described through these vocabularies, will be thus accessible as organized with explicit semantic references for humans as well as for machines. Yet, to be useful, ontologies should cover large part of human knowledge. Automatically learning these ontologies from document collections is the major challenge. Models for automatically learning semantic networks of words from texts use both corpus-extracted evidences and existing language resources (Basili, Gliozzo, & Pennacchiotti, 2007). All these models rely on two hypotheses: Distributional Hypothesis (DH) (Harris, 1964) and Lexico-syntactic patterns exploitation hypothesis (LSP) (Robison, 1970). While these are powerful tools to extract relations among concepts using texts, models based on these hypotheses do not explicitly exploit structural properties of target relations when learning taxonomies or semantic networks of words. DH models intrinsically use structural properties of semantic networks of words such as transitivity, but these models cannot be applied for l...

show abstract

Model adaptation via model interpolation and boosting for web search ranking

Cited by 22 publications

References 18 publications

Ranking function adaptation with boosting trees

Ranking function adaptation with boosting trees

Fractional Similarity: Cross-Lingual Feature Selection for Search

Exploiting Transitivity in Probabilistic Models for Ontology Learning

Contact Info

Product

Resources

About