Min-Yen Kan

Question answering passage retrieval using dependency relations

State-of-the-art question answering (QA) systems employ term-density ranking to retrieve answer passages. Such methods often retrieve incorrect passages as relationships among question terms are not considered. Previous studies attempted to address this problem by matching dependency relations between questions and answers. They used strict matching, which fails when semantically equivalent relationships are phrased differently. We propose fuzzy relation matching based on statistical models. We present two methods for learning relation mapping scores from past QA pairs: one based on mutual information and the other on expectation maximization. Experimental results show that our method significantly outperforms state-of-the-art density-based passage retrieval methods by up to 78% in mean reciprocal rank. Relation matching also brings about a 50% improvement in a system enhanced by query expansion.

A PDTB-Styled End-to-End Discourse Parser

Since the release of the large discourse-level annotation of the Penn Discourse Treebank (PDTB), research work has been carried out on certain subtasks of this annotation, such as disambiguating discourse connectives and classifying Explicit or Implicit relations. We see a need to construct a full parser on top of these subtasks and propose a way to evaluate the parser. In this work, we have designed and developed an end-to-end discourse parser-to-parse free texts in the PDTB style in a fully data-driven approach. The parser consists of multiple components joined in a sequential pipeline architecture, which includes a connective classifier, argument labeler, explicit classifier, non-explicit classifier, and attribution span labeler. Our trained parser first identifies all discourse and non-discourse relations, locates and labels their arguments, and then classifies the sense of the relation between each pair of arguments. For the identified relations, the parser also determines the attribution spans, if any, associated with them. We introduce novel approaches to locate and label arguments, and to identify attribution spans. We also significantly improve on the current state-of-the-art connective classifier. We propose and present a comprehensive evaluation from both component-wise and error-cascading perspectives, in which we illustrate how each component performs in isolation, as well as how the pipeline performs with errors propagated forward. The parser gives an overall system F 1 score of 46.80 percent for partial matching utilizing gold standard parses, and 38.18 percent with full automation.

Fast webpage classification using URL features

We demonstrate the usefulness of the uniform resource locator (URL) alone in performing web page classification. This approach is faster than typical web page classification, as the pages do not have to be fetched and analyzed. Our approach segments the URL into meaningful chunks and adds component, sequential and orthographic features to model salient patterns. The resulting features are used in supervised maximum entropy modeling. We analyze our approachs effectiveness on two standardized domains. Our results show that in certain scenarios, URL-based methods approach the performance of current state-of-the-art full-text and link-based methods.

Simfinder: A flexible clustering tool for summarization

Abstract : We present a statistical similarity measuring and clustering tool, SIMFINDER, that organizes small pieces of text from one or multiple documents into tight clusters. By placing highly related text units in the same cluster, SIMFINDER enables a subsequent content selection/generation component to reduce each cluster to a single sentence, either by extraction or by reformulation. We report on improvements in the similarity and clustering components of SIMFINDER, including a quantitative evaluation, and establish the generality of the approach by interfacing SIMFINDER to two very different summarization systems.

Fast Matrix Factorization for Online Recommendation with Implicit Feedback

This paper contributes improvements on both the effectiveness and efficiency of Matrix Factorization (MF) methods for implicit feedback. We highlight two critical issues of existing works. First, due to the large space of unobserved feedback, most existing works resort to assign a uniform weight to the missing data to reduce computational complexity. However, such a uniform assumption is invalid in real-world settings. Second, most methods are also designed in an offline setting and fail to keep up with the dynamic nature of online data. We address the above two issues in learning MF models from implicit feedback. We first propose to weight the missing data based on item popularity, which is more effective and flexible than the uniform-weight assumption. However, such a non-uniform weighting poses efficiency challenge in learning the model. To address this, we specifically design a new learning algorithm based on the element-wise Alternating Least Squares (eALS) technique, for efficiently optimizing a MF model with variably-weighted missing data. We exploit this efficiency to then seamlessly devise an incremental update strategy that instantly refreshes a MF model given new feedback. Through comprehensive experiments on two public datasets in both offline and online protocols, we show that our implemented, open-source (https://github.com/hexiangnan/sigir16-eals) eALS consistently outperforms state-of-the-art implicit MF methods.

Keyphrase extraction in scientific publications

We present a keyphrase extraction algorithm for scientific publications. Different from previous work, we introduce features that capture the positions of phrases in document with respect to logical sections found in scientific discourse. We also introduce features that capture salient morphological phenomena found in scientific keyphrases, such as whether a candidate keyphrase is an acronyms or uses specific terminologically productive suffixes. We have implemented these features on top of a baseline feature set used by Kea [1]. In our evaluation using a corpus of 120 scientific publications multiply annotated for keyphrases, our system significantly outperformed Kea at the p < .05 level. As we know of no other existing multiply annotated keyphrase document collections, we have also made our evaluation corpus publicly available. We hope that this contribution will spur future comparative research.

Search engine driven author disambiguation

In scholarly digital libraries, author disambiguation is an important task that attributes a scholarly work with specific authors. This is critical when individuals share the same name. We present an approach to this task that analyzes the results of automatically-crafted Web searches. A key observation is that pages from rare Web sites are stronger source of evidence than pages from common Web sites, which we model as inverse host frequency (IHF). Our system is able to achieve an average accuracy of 0.836

Addressing cold-start in app recommendation: latent user models constructed from twitter followers

As a tremendous number of mobile applications (apps) are readily available, users have difficulty in identifying apps that are relevant to their interests. Recommender systems that depend on previous user ratings (i.e., collaborative filtering, or CF) can address this problem for apps that have sufficient ratings from past users. But for apps that are newly released, CF does not have any user ratings to base recommendations on, which leads to the cold-start problem. In this paper, we describe a method that accounts for nascent information culled from Twitter to provide relevant recommendation in such cold-start situations. We use Twitter handles to access an apps Twitter account and extract the IDs of their Twitter-followers. We create pseudo-documents that contain the IDs of Twitter users interested in an app and then apply latent Dirichlet allocation to generate latent groups. At test time, a target user seeking recommendations is mapped to these latent groups. By using the transitive relationship of latent groups to apps, we estimate the probability of the user liking the app. We show that by incorporating information from Twitter, our approach overcomes the difficulty of cold-start app recommendation and significantly outperforms other state-of-the-art recommendation techniques by up to 33%.

Paraphrase Recognition via Dissimilarity Significance Classification

We propose a supervised, two-phase framework to address the problem of paraphrase recognition (PR). Unlike most PR systems that focus on sentence similarity, our framework detects dissimilarities between sentences and makes its paraphrase judgment based on the significance of such dissimilarities. The ability to differentiate significant dissimilarities not only reveals what makes two sentences a non-paraphrase, but also helps to recall additional paraphrases that contain extra but insignificant information. Experimental results show that while being accurate at discerning non-paraphrasing dissimilarities, our implemented system is able to achieve higher paraphrase recall (93%), at an overall performance comparable to the alternatives.

Adaptive sorted neighborhood methods for efficient record linkage

Traditionally, record linkage algorithms have played an important role in maintaining digital libraries - i.e., identifying matching citations or authors for consolidation in updating or integrating digital libraries. As such, a variety of record linkage algorithms have been developed and deployed successfully. Often, however, existing solutions have a set of parameters whose values are set by human experts off-lineand are fixed during the execution. Since finding the ideal values of such parameters is not straightforward, or no such single ideal value even exists, the applicability of existing solutions to new scenarios or domains is greatly hampered. To remedy this problem, we argue that one can achieve significant improvement by adaptively and dynamically changing such parameters of record linkage algorithms. To validate our hypothesis, we take a classical record linkage algorithm, the sorted neighborhood method (SNM), and demonstrate how we can achieve improved accuracy and performance by adaptively changing its fixed sliding window size. Our claim is analytically and empirically validated using both real and synthetic data sets of digital libraries and other domains.