Dialog state tracking (DST), which infers user goals in the presence of noise, is important for spoken dialog systems. Recently it has attracted a lot attention research community. Several new approaches have been proposed, especially series DST Challenges (DSTC). But problem cross-domain generalization, i.e., whether trackers designed one domain will perform similarly well on other domains, still an open issue. This becomes focus DSTC3. To tackle this problem, we adopt domain-independent...

Traditional supervised learning approaches to common NLP tasks depend heavily on manual annotation, which is labor intensive and time consuming, often suffer from data sparseness.In this paper we show how mitigate the problems in short text classification (STC) through word embeddings -distributional representations of words learned large unlabeled data.The are trained entire English Wikipedia text.We assume that a document specific sample one distribution Bayesian framework.A Gaussian...

Type 2 Diabetes Mellitus (T2DM) is a chronic disease closely related to personal life style. Therefore, achieving effective self-management one of the most important ways control it. There evidence that social support can help improve ability patients with T2DM, but which more has been rarely explored. The purpose this study construct an integrated model analyze significant impact on and provide reasonable suggestions health care providers how effectively play role support.We established...

Appropriate location is an important prerequisite for the long-term survival and development of private medical institutions. However, in both theory practice, issue decision-making clinics has not been fully studied. We therefore aimed to provide a feasible scheme new clinics. This paper combines k-means clustering method integrated 2DULVs (two-dimensional uncertain language variables)-TOPSIS (technique order preference by similarity ideal solution)-DSCCR (Dempster-Shafer conjunctive...

Discriminative dialog state tracking has become a hot topic in research community recently.Compared to generative approach, it the advantage of being able handle arbitrary dependent features, which is very appealing.In this paper, we present our approach DSTC2 challenge.We propose use discriminative Markovian models as natural enhancement stationary models.The structure allows incorporation 'transitional' can lead more efficiency and flexibility user goal changes.Results on dataset show...

In this paper, we propose a method to predict the user emotional state (anger or neutral) for improvement of satisfaction in call center. order detect more accurately, our work employs following information fusion technologies: (1) view data imbalance problem, adopt statistical model fusion, (2) improving classifier performance, combine features with n-gram, sentiment word and domain-specific word, (3) according characteristics spoken language, use combination language rule, (4) as multi...

This paper describes our systems for expression-level and message-level sentiment analysis -two subtasks of SemEval-2015 Task 10 on in Twitter.First we built two baseline the using SVM with a variety features.Then improved through model iteration probability-output weighting respectively.Our submissions are ranked 3rd 2nd among eleven teams 2015 test set progress subtask A 7th 4th 40 sets respectively B.

There have been considerable attempts to incorporate semantic knowledge into coreference resolution systems: different sources such as WordNet and Wikipedia used boost the performance. In this paper, we propose new ways extract feature. This feature, along with other features named entity can be build an accurate class (SC) classifier. addition, analyze SC classification errors use relaxed agreement features. The proposed classifier relaxation of on ACE2 evaluation our baseline system by...

Automatic optimization of spoken dialog management policies that are robust to environmental noise has long been the goal for both academia and industry. Approaches based on reinforcement learning have proved be effective. However, numerical representation policy is human-incomprehensible difficult system designers verify or modify, which limits its practical application. In this paper we propose a novel framework optimizing specified in domain language using genetic algorithm. The...

Abstract In China, Type 2 diabetes mellitus (T2DM) is increasingly affecting people's health. Although many risk factors related to T2DM have been researched, the association between social relationships and management of in China has not fully researched. Therefore, we obtained 2,969 valid cases from National Chinese Medicine Clinical Research Base‐Key Disease Diabetes Mellitus Study evaluate role T2DM. We first establish an indicators system relationship then propose a comprehensive method...

This paper describes our system for Shallow Discourse Parsing -the CoNLL 2015 Shared Task.We regard this as a classification task and build cascaded based on Maximum Entropy to identify the discourse connective, spans of two arguments sense connective.We trained models with variety features such lexical syntactic features.We also report results achieved by team.

Voice search is the technology that enables users to access information using spoken queries. Automatic speech recognizer (ASR) one of key modules for voice systems. However, high error rate state-of-the-art large vocabulary continuous recognition (LVCSR) bottleneck most In this paper, we first build a baseline system language model (LM) with domain-specific information. To improve our system, propose forward-backward LVCSR combination method decrease errors in recognition. Experiment...

Abstract Face swiping is the action of registering facial recognition data with a biometric authentication application. swiping‐based medical services (FSMSs) an emerging measurement technology underpinning modern intelligent in China. In view convenience delivery and potential to reduce costs, FSMSs have become important part China's reform. However, modelling its eventual uses has not been fully explored theory practice, this hindering take up paper, we build evolutionary game model...

Robustness is one of the most challenging issues for spoken language understanding (SLU). In this paper we studied semantic Chinese a voice search dialogue system. We first simplified problem into named entity recognition (NER) task, which was further formulated as sequential tagging. carried out experiments to opt character over word tagging unit. Then two approaches were proposed exploit prior knowledge - in form domain lexicon character-based framework. One enriched tagger features by...

In this paper we describe our work on generating in-domain corpus using auto-induced semantic classes and structures for language model adaptation in a voice search dialogue system. We proposed novel similarity measure based co-occurrence probabilities inducing classes. Clustering with the new outperformed that widely used distance Kullback-Leibler divergence. For adaptation, adopted approach of interpolation. Experiments show both human-human generated data helped lot latter more. This...

Abstract Background: In the context of "Internet +" medical treatment, mobile health applications provide services for people in a new way, making it possible to carry out management anytime and anywhere. According survey data, most powerful consumers field are those aged 24 35. Thus, can be seen, is particularly important study preferences young applications.Methods: This established domain-adaptive application evaluation model based on users’ experience, used an interactive algorithm...