A neural conversational model (NCM) based on an encoder-decoder recurrent neural network (RNN) with an attention mechanism learns different sequence-to-sequence mappings from what neural machine translation (NMT) learns even when based on the same technique. In the NCM, we confirmed that target-word-to-source-word mappings captured by the attention mechanism are not as clear and stationary as those for NMT. Considering that vector norms indicate a magnitude of information in the processing, we analyzed the inner workings of an encoder-decoder GRU-based NCM focusing on the norms of word embedding vectors and hidden vectors. First, we conducted correlation analyses on the norms of word embedding vectors with frequencies in the training set and with conditional entropies of a bi-gram language model to understand what is correlated with the norms in the encoder and decoder. Second, we conducted correlation analyses on norms of change in the hidden vector of the recurrent layer with their input vectors for the encoder and decoder, respectively. These analyses were done to understand how the magnitude of information propagates through the network. The analytical results suggested that the norms of the word embedding vectors are associated with their semantic information in the encoder, while those are associated with the predictability as a language model in the decoder. The analytical results further revealed how the norms propagate through the recurrent layer in the encoder and decoder.

Query-by-example spoken term detection (QbE-STD) is a task of using speech queries to match utterances, and the acoustic word embedding (AWE) method of generating fixed-length representations for speech segments has shown high performance and efficiency in recent work. We propose an AWE training method using a label-adversarial network to reduce the interference information learned during AWE training. Experiments demonstrate that our method achieves significant improvements on multilingual and zero-resource test sets.

To solve the low accuracy problem of the recommender system for long term users, in this paper, we propose a top-N-balanced sequential recommendation based on recurrent neural network. We postulated and verified that the interactions between users and items is time-dependent in the long term, but in the short term, it is time-independent. We balance the top-N recommendation and sequential recommendation to generate a better recommender list by improving the loss function and generation method. The experimental results demonstrate the effectiveness of our method. Compared with a state-of-the-art recommender algorithm, our method clearly improves the performance of the recommendation on hit rate. Besides the improvement of the basic performance, our method can also handle the cold start problem and supply new users with the same quality of service as the old users.

A tremendous amount of knowledge is present in the ever-growing scientific literature. In order to efficiently grasp such knowledge, various computational tasks are proposed that train machines to read and analyze scientific documents. One of these tasks, Scientific Relation Extraction, aims at automatically capturing scientific semantic relationships among entities in scientific documents. Conventionally, only a limited number of commonly used knowledge bases, such as Wikipedia, are used as a source of background knowledge for relation extraction. In this work, we hypothesize that unannotated scientific papers could also be utilized as a source of external background information for relation extraction. Based on our hypothesis, we propose a model that is capable of extracting background information from unannotated scientific papers. Our experiments on the RANIS corpus [1] prove the effectiveness of the proposed model on relation extraction from scientific articles.

In this paper, we propose a deep learning based model for classifying speech-acts using a convolutional neural network (CNN). The model uses some bigram features including parts-of-speech (POS) tags and dependency-relation bigrams, which represent syntactic structural information in utterances. Previous classification approaches using CNN have commonly exploited word embeddings using morpheme unigrams. However, the proposed model first extracts two different bigram features that well reflect the syntactic structure of utterances and then represents them as a vector representation using a word embedding technique. As a result, the proposed model using bigram embeddings achieves an accuracy of 89.05%. Furthermore, the accuracy of this model is relatively 2.8% higher than that of competitive models in previous studies.

Web search queries are usually vague, ambiguous, or tend to have multiple intents. Users have different search intents while issuing the same query. Understanding the intents through mining subtopics underlying a query has gained much interest in recent years. Query suggestions provided by search engines hold some intents of the original query, however, suggested queries are often noisy and contain a group of alternative queries with similar meaning. Therefore, identifying the subtopics covering possible intents behind a query is a formidable task. Moreover, both the query and subtopics are short in length, it is challenging to estimate the similarity between a pair of short texts and rank them accordingly. In this paper, we propose a method for mining and ranking subtopics where we introduce multiple semantic and content-aware features, a bipartite graph-based ranking (BGR) method, and a similarity function for short texts. Given a query, we aggregate the suggested queries from search engines as candidate subtopics and estimate the relevance of them with the given query based on word embedding and content-aware features by modeling a bipartite graph. To estimate the similarity between two short texts, we propose a Jensen-Shannon divergence based similarity function through the probability distributions of the terms in the top retrieved documents from a search engine. A diversified ranked list of subtopics covering possible intents of a query is assembled by balancing the relevance and novelty. We experimented and evaluated our method on the NTCIR-10 INTENT-2 and NTCIR-12 IMINE-2 subtopic mining test collections. Our proposed method outperforms the baselines, known related methods, and the official participants of the INTENT-2 and IMINE-2 competitions.

Building high-quality text-to-speech (TTS) systems without expert knowledge of the target language and/or time-consuming manual annotation of speech and text data is an important yet challenging research topic. In this kind of TTS system, it is vital to find representation of the input text that is both effective and easy to acquire. Recently, the continuous representation of raw word inputs, called “word embedding”, has been successfully used in various natural language processing tasks. It has also been used as the additional or alternative linguistic input features to a neural-network-based acoustic model for TTS systems. In this paper, we further investigate the use of this embedding technique to represent phonemes, syllables and phrases for the acoustic model based on the recurrent and feed-forward neural network. Results of the experiments show that most of these continuous representations cannot significantly improve the system's performance when they are fed into the acoustic model either as additional component or as a replacement of the conventional prosodic context. However, subjective evaluation shows that the continuous representation of phrases can achieve significant improvement when it is combined with the prosodic context as input to the acoustic model based on the feed-forward neural network.

Short texts usually encounter the problem of data sparseness, as they do not provide sufficient term co-occurrence information. In this paper, we show how to mitigate the problem in short text classification through word embeddings. We assume that a short text document is a specific sample of one distribution in a Gaussian-Bayesian framework. Furthermore, a fast clustering algorithm is utilized to expand and enrich the context of short text in embedding space. This approach is compared with those based on the classical bag-of-words approaches and neural network based methods. Experimental results validate the effectiveness of the proposed method.