Ranking Model Adaptation for Domain-Specific Search

With the explosive emergence of vertical search domains, applying the broad-based ranking model directly to different domains is no longer desirable due to domain differences, while building a unique ranking model for each domain is both laborious for labeling data and time-consuming for training models. In this paper, we address these difficulties by proposing a regularization based algorithm called ranking adaptation SVM (RA-SVM), through which we can adapt an existing ranking model to a new domain, so that the amount of labeled data and the training cost is reduced while the performance is still guaranteed.

Our algorithm only requires the Prediction from the existing ranking models, rather than their internal representations or the data from auxiliary domains. In addition, we assume that documents similar in the domain-specific feature space should have consistent rankings, and add some constraints to control the margin and slack variables of RA-SVM adaptively. Finally, ranking adaptability measurement is proposed to quantitatively estimate if an existing ranking model can be adapted to a new domain. Experiments performed over Letor and two large scale datasets crawled from a commercial search engine demonstrate the applicabilities of the proposed ranking adaptation algorithms and the ranking adaptability measurement.

Existing System:

The existing broad-based ranking model provides a lot of common information in ranking documents only few training samples are needed to be labeled in the new domain. From the probabilistic perspective, the broad-based ranking model provides a prior knowledge, so that only a small number of labeled samples are sufficient for the target domain ranking model to achieve the same confidence. Hence, to reduce the cost for new verticals, how to adapt the auxiliary ranking models to the new target domain and make full use of their domain-specific features, turns into a pivotal problem for building effective domain-specific ranking models.

Proposed System:

Proposed System focus whether we can adapt ranking models learned for the existing broad-based search or some verticals, to a new domain, so that the amount of labeled data in the target domain is reduced while the performance requirement is still guaranteed, how to adapt the ranking model effectively and efficiently and how to utilize domain-specific features to further boost the model adaptation. The first problem is solved by the proposed rank-ing adaptability measure, which quantitatively estimates whether an existing ranking model can be adapted to the new domain, and predicts the potential performance for the adaptation.
We address the second problem from the regularization framework and a ranking adaptation SVM algorithm is proposed. Our algorithm is a blackbox ranking model adaptation, which needs only the predictions from the existing ranking model, rather than the internal representation of the model itself or the data from the auxiliary domains. With the black-box adaptation property, we achieved not only the flexibility but also the efficiency. To resolve the third problem, we assume that documents similar in their domain specific feature space should have consistent rankings.

Modules:

  • Ranking Adaptation Module
  • Explore Ranking adaptability Module
  • Ranking adaptation with domain specific search Module
  • Ranking Support Vector Machine Module

Tools Used:

Front End : JAVA and J2EE