Inho Kang

Bio: Inho Kang is an academic researcher from Naver Corporation. The author has contributed to research in topics: Computer science & Sentence. The author has an hindex of 10, co-authored 27 publications receiving 984 citations. Previous affiliations of Inho Kang include KAIST & Samsung.

Query type classification for web document retrieval

Abstract: The heterogeneous Web exacerbates IR problems and short user queries make them worse. The contents of web documents are not enough to find good answer documents. Link information and URL information compensates for the insufficiencies of content information. However, static combination of multiple evidences may lower the retrieval performance. We need different strategies to find target documents according to a query type. We can classify user queries as three categories, the topic relevance task, the homepage finding task, and the service finding task. In this paper, a user query classification scheme is proposed. This scheme uses the difference of distribution, mutual information, the usage rate as anchor texts, and the POS information for the classification. After we classified a user query, we apply different algorithms and information for the better results. For the topic relevance task, we emphasize the content information, on the other hand, for the homepage finding task, we emphasize the Link information and the URL information. We could get the best performance when our proposed classification method with the OKAPI scoring algorithm was used.

Apparatus for providing voice dialogue service and method of operating the same

Abstract: A speech dialogue service apparatus including: a language analysis module tagging a part of speech (POS) of each respective word included in a sentence recorded in a predetermined text, syntactically analyzing the sentence by classifying a meaning of each respective word, and generating at least one semantic frame corresponding to the sentence according to a result of the syntactical analysis; and a dialogue management module analyzing an intention of the sentence corresponding to the at least one respective semantic frame, and generating a system response corresponding to the sentence intention by selecting a predetermined sentence intention according to whether an action corresponding to the intention of the respective sentence can be performed.

Semantic Sentence Matching with Densely-connected Recurrent and Co-attentive Information

Abstract: Sentence matching is widely used in various natural language tasks such as natural language inference, paraphrase identification, and question answering. For these tasks, understanding logical and semantic relationship between two sentences is required but it is yet challenging. Although attention mechanism is useful to capture the semantic relationship and to properly align the elements of two sentences, previous methods of attention mechanism simply use a summation operation which does not retain original features enough. Inspired by DenseNet, a densely connected convolutional network, we propose a densely-connected co-attentive recurrent neural network, each layer of which uses concatenated information of attentive features as well as hidden features of all the preceding recurrent layers. It enables preserving the original and the co-attentive feature information from the bottommost word embedding layer to the uppermost recurrent layer. To alleviate the problem of an ever-increasing size of feature vectors due to dense concatenation operations, we also propose to use an autoencoder after dense concatenation. We evaluate our proposed architecture on highly competitive benchmark datasets related to sentence matching. Experimental results show that our architecture, which retains recurrent and attentive features, achieves state-of-the-art performances for most of the tasks.

Apparatus and method for detecting named entity

Abstract: An apparatus and method for detecting a named-entity. The apparatus includes a candidate-named-entity extraction module that detects a candidate-named-entity based on an initial learning example and feature information regarding morphemes constituting an inputted sentence, the candidate-named-entity extraction module providing a tagged sentence including the detected candidate-named-entity; a storage module that stores information regarding a named-entity dictionary and a rule; and a learning-example-regeneration module for finally determining whether the candidate-named-entity included in the provided sentence is a valid named-entity, based on the named-entity dictionary and the rule, the learning-example-regeneration module providing the sentence as a learning example, based on a determination result, so that a probability of candidate-named-entity detection is gradually updated.

Semantic Sentence Matching with Densely-connected Recurrent and Co-attentive Information

Abstract: Sentence matching is widely used in various natural language tasks such as natural language inference, paraphrase identification, and question answering. For these tasks, understanding logical and semantic relationship between two sentences is required but it is yet challenging. Although attention mechanism is useful to capture the semantic relationship and to properly align the elements of two sentences, previous methods of attention mechanism simply use a summation operation which does not retain original features enough. Inspired by DenseNet, a densely connected convolutional network, we propose a densely-connected co-attentive recurrent neural network, each layer of which uses concatenated information of attentive features as well as hidden features of all the preceding recurrent layers. It enables preserving the original and the co-attentive feature information from the bottommost word embedding layer to the uppermost recurrent layer. To alleviate the problem of an ever-increasing size of feature vectors due to dense concatenation operations, we also propose to use an autoencoder after dense concatenation. We evaluate our proposed architecture on highly competitive benchmark datasets related to sentence matching. Experimental results show that our architecture, which retains recurrent and attentive features, achieves state-of-the-art performances for most of the tasks.

Learning to Rank for Information Retrieval

Abstract: This tutorial is concerned with a comprehensive introduction to the research area of learning to rank for information retrieval. In the first part of the tutorial, we will introduce three major approaches to learning to rank, i.e., the pointwise, pairwise, and listwise approaches, analyze the relationship between the loss functions used in these approaches and the widely-used IR evaluation measures, evaluate the performance of these approaches on the LETOR benchmark datasets, and demonstrate how to use these approaches to solve real ranking applications. In the second part of the tutorial, we will discuss some advanced topics regarding learning to rank, such as relational ranking, diverse ranking, semi-supervised ranking, transfer ranking, query-dependent ranking, and training data preprocessing. In the third part, we will briefly mention the recent advances on statistical learning theory for ranking, which explain the generalization ability and statistical consistency of different ranking methods. In the last part, we will conclude the tutorial and show several future research directions.

Intelligent Automated Assistant

Abstract: An intelligent automated assistant system engages with the user in an integrated, conversational manner using natural language dialog, and invokes external services when appropriate to obtain information or perform various actions. The system can be implemented using any of a number of different platforms, such as the web, email, smartphone, and the like, or any combination thereof. In one embodiment, the system is based on sets of interrelated domains and tasks, and employs additional functionally powered by external services with which the system can interact.

Automated Response to and Sensing of User Activity in Portable Devices

Abstract: The various methods and devices described herein relate to devices which, in at least certain embodiments, may include one or more sensors for providing data relating to user activity and at least one processor for causing the device to respond based on the user activity which was determined, at least in part, through the sensors. The response by the device may include a change of state of the device, and the response may be automatically performed after the user activity is determined.

Translation techniques in cross-language information retrieval

Abstract: Cross-language information retrieval (CLIR) is an active sub-domain of information retrieval (IR). Like IR, CLIR is centered on the search for documents and for information contained within those documents. Unlike IR, CLIR must reconcile queries and documents that are written in different languages. The usual solution to this mismatch involves translating the query and/or the documents before performing the search. Translation is therefore a pivotal activity for CLIR engines. Over the last 15 years, the CLIR community has developed a wide range of techniques and models supporting free text translation. This article presents an overview of those techniques, with a special emphasis on recent developments.

Multi-Task Deep Neural Networks for Natural Language Understanding

Abstract: In this paper, we present a Multi-Task Deep Neural Network (MT-DNN) for learning representations across multiple natural language understanding (NLU) tasks. MT-DNN not only leverages large amounts of cross-task data, but also benefits from a regularization effect that leads to more general representations to help adapt to new tasks and domains. MT-DNN extends the model proposed in Liu et al. (2015) by incorporating a pre-trained bidirectional transformer language model, known as BERT (Devlin et al., 2018). MT-DNN obtains new state-of-the-art results on ten NLU tasks, including SNLI, SciTail, and eight out of nine GLUE tasks, pushing the GLUE benchmark to 82.7% (2.2% absolute improvement) as of February 25, 2019 on the latest GLUE test set. We also demonstrate using the SNLI and SciTail datasets that the representations learned by MT-DNN allow domain adaptation with substantially fewer in-domain labels than the pre-trained BERT representations. Our code and pre-trained models will be made publicly available.