Search

Sentiment classification aims to detect polarity from a piece of text. The polarity is usually positive or negative, and the text genre is usually product review. The challenges of sentiment classification are that it is hard to capture semantic of reviews, and the labeled data is hard to annotate. Therefore, we propose neural co-training to learn the semantic representation of each review using the neural network model, and learn the information from unlabeled data using a co-training framework. In particular, we use the attention-based bi-directional Gated Recurrent Unit (Att-BiGRU) to model the semantic content of each review and regard different categories of the target product as different views. We then use a co-training framework to learn and predict the unlabeled reviews with different views. Experiment results with the Yelp dataset demonstrate the effectiveness of our approach.

In social networks, rising stars are junior individuals who may be not so charming at first but turn out to be outstanding over time. Recently, rising star evaluation has become a popular research topic in the field of social analysis, which is helpful for decision support, cognitive computation, and other practical problems. In this paper, we study the problem of rising star evaluation in geo-social networks. Specifically, given a topic keyword Q and a time point t, we aim at evaluating the latent influence of users to find rising stars, which refer to experts who have few activities and little impact currently on the underlying geo-social network but may become influential experts in the future. To efficiently evaluate future stars, we propose a novel processing framework based on extreme learning machine (ELM) called FS-ELM. FS-ELM consists of three key components. The first component constructs features by incorporating social topology and user behavior patterns. The second component extracts supervised information by discovering topic experts of Q at time (t + Δt); that is, excluding those detected at time t, topic experts obtained at time (t + Δt) can be regarded as rising stars at time t. The third component is ELM-based future star classification that leverages ELM as a departure point to evaluate whether a user is a rising star. Our experimental studies conducted on real-world datasets show that (1) FS-ELM can effectively discover rising stars with a query topic at time t and outperform other traditional methods and (2) user social characteristics have an important impact on the rising star evaluation. This paper studies a novel problem, namely, rising star evaluation in geo-social networks. We propose an advanced processing framework based on ELM by exploiting social topology characteristics and user behavior patterns. The experimental results encouragingly demonstrate the efficiency and effectiveness of the proposed approach.

Event Argument Extraction aims to extract various arguments of a specific event and assign the roles to them. Currently, most of previous work focused on sentence-level task, ignoring the whole contexts of the event mentions. Obviously, the event mentions in a discourse always are around a specific topic and have interior relations, especially those intra-sentence or adjacent-sentence event mentions. Therefore, this paper proposes an event argument extraction graph model on discourse level. It adds various event mentions, entity mentions and partial semantic dependencies into a graph, which aggregates the discourse contexts by event information, entity information and semantic dependency. The experimental results on the ACE2005 Chinese corpus show that our proposed model outperforms several strong baselines.

Graph similarity search is an important research problem in many applications, such as finding result graphs that have a similar structure to a given entity in biochemistry, data mining, and pattern recognition. Top-k graph similarity search is one of graph similarity search tasks, which aims to find the top-k graphs that are most similar to the query graph in a given graph database. In this paper, the top-k similarity search problem based on the graph edit distance is studied according to the corollary of the partitioned similarity theorem. Firstly, in order to speed up the online search process and avoid scanning each graph in the database one by one, an offline hierarchical inverted index is constructed to satisfy top-k search. Secondly, the offline hierarchical inverted index is used to filter the candidate graphs online and verify them, which reduces the time of searching the graphs. Finally, the good performance of the algorithm in running time and scalability is verified by running the similarity algorithm on real dataset and synthetic dataset.

Survey generation aims to generate a summary from a scientific topic based on related papers. The structure of papers deeply influences the generative process of survey, especially the relationships between sentence and sentence, paragraph and paragraph. In principle, the structure of paper can influence the quality of the summary. Therefore, we employ the structure of paper to leverage contextual information among sentences in paragraphs to generate a survey for documents. In particular, we present a neural document structure model for survey generation. We take paragraphs as units, and model sentences in paragraphs, we then employ a hierarchical model to learn structure among sentences, which can be used to select important and informative sentences to generate survey. We evaluate our model on scientific document data set. The experimental results show that our model is effective, and the generated survey is informative and readable.

Stance detection aims to automatically determine whether the author is in favor of or against a given target. In principle, the sentiment information of a post highly influences the stance. In this study, we aim to leverage the sentiment information of a post to improve the performance of stance detection. However, conventional discrete models with sentimental features can cause error propagation. We thus propose a joint neural network model to predict the stance and sentiment of a post simultaneously, because the neural network model can learn both representation and interaction between the stance and sentiment collectively. Specifically, we first learn a deep shared representation between stance and sentiment information, and then use a neural stacking model to leverage sentimental information for the stance detection task. Empirical studies demonstrate the effectiveness of our proposed joint neural model.

Previous research on emotions analysis has placed much emphasis in monolingual instead of bilingual text. However, emotions on social media platforms are often found in bilingual or code-switching posts. Different from monolingual text, emotions in code-switching text can be expressed in both monolingual and bilingual forms. Moreover, more than one emotion can be expressed within a single post; yet they tend to be related in some ways which offers some implications. It is thus necessary to consider the correlation between different emotions. In this paper, a joint factor graph model is proposed to address this issue. In particular, attribute functions of the factor graph model are utilized to learn both monolingual and bilingual information from each post, factor functions are used to explore the relationship among different emotions, and a belief propagation algorithm is employed to learn and predict the model. Empirical studies demonstrate the importance of emotion analysis in code-switching text and the effectiveness of our proposed joint learning model.

Event detection is the basic task of event extraction. Previous studies usually used independent sentences as basic event detection objects. They cannot effectively identify event triggers which depend on document information. Besides, there are correlations between the sentences and words in the document. Therefore, it is necessary to use document information for event detection. In this study, we propose a graph model for event detection based on document structure. It is used to connect sentences and words in a document. Specifically, we finetune BERT model and use Bi-LSTM to learn the sentences and their context features, and then use GCN to model the document relation graph. The document relation graph is based on the parts of speech of all words in different sentences, which contributes to establishing the triggers-triggers relation and triggers-arguments relation. The experimental results on LitBank show that our proposed model outperforms all baselines significantly and verifies the validity of document information.

Sentiment forecasting in dialog aims to predict the polarity of next utterance to come, and can help speakers revise their utterances in sentimental utterances generation. However, the polarity of next utterance is normally hard to predict, due to the lack of content of next utterance (yet to come). In this study, we propose a Neural Sentiment Forecasting (NSF) model to address inherent challenges. In particular, we employ a neural simulation model to simulate the next utterance based on the context (previous utterances encountered). Moreover, we employ a sequence influence model to learn both pair-wise and seq-wise influence. Empirical studies illustrate the importance of proposed sentiment forecasting task, and justify the effectiveness of our NSF model over several strong baselines.

Question Generation (QG) aims to generate a question based on the context. Given the intrinsic connections between QG and QA (Question Answering), we focus on training a joint model for both QG and QA, and take one step further to integrate one more context self-encoding (CSE) task into the joint model, as a junction auxiliary task to better integrate QG and QA. In particular, our model employs a cross-task autoencoder to incorporate QG, QA and CSE into a joint learning process, which could better utilize the correlation between the contexts of different tasks in learning representations and provide more task-specific information. Experimental results show the effectiveness of our triple-task training model for QG, and the importance of learning interaction among QA and CSE for QG.

With the continuous development of the Internet, the field of e-commerce generates many comments on products. It is of great significance for both merchants and customers to generate product-related questions by utilizing a large amount of review information of products. In order to get rid of the traditional constraints of generating models based on artificial rules and make the question generation more accurate, this paper proposes a question generation model based on product information. Compared with the existing approaches, this model can generate questions more relevant to the products, and more fluent. In particular, the model can not only avoid the problem that the vocabulary exceeds the dictionary, but also extract the vocabulary needed for question generation from the original text and the dictionary. The experimental results show that in the task of generating short text based on comments, compared with the existing neural network model, the effectiveness has been greatly improved.

This paper presents the overview of the shared task, emotion detection in code-switching text, in NLPCC 2018. The submitted systems are expected to automatically determine the emotions in the Chinese-English code-switching text. Different from monolingual text, code-switching text contains more than one language, and the emotion can be expressed by either monolingual or bilingual form. Hence, the challenges are: how to integrate both monolingual and bilingual forms to detect emotion, and how to bridge the gap to between two languages. Our shared task has 19 team participants. The highest F-score was 0.515. In this paper, we introduce the task, the corpus, the participating teams, and the evaluation results.

In the literature, various supervised learning approaches have been adopted to address the task of emotion classification. However, the performance of these approaches greatly suffers when the size of the labeled data is limited. In this paper, we tackle this challenge from a cross-lingual sensoria where the labeled data in a resource-rich language (i.e., English in this study) is employed to improve the emotion classification performance in a resource-poor language (i.e., Chinese in this study). Specifically, we first use machine translation services to eliminate the language gap between Chinese and English data and then propose a joint learning framework to leverage both Chinese and English data, which develops auxiliary representations from several auxiliary emotion classification tasks. Furthermore, in our joint learning approach, we introduce an attention mechanism to capture informative words. Empirical studies demonstrate the effectiveness of the proposed approach to emotion classification.

Distributed Denial of Service (DDoS) attacks have been significant threats to the Internet. Traditional research in cyber security focuses on detecting emerging DDoS attacks by tracing network package flow. A characteristic of DDoS defense is that rescue time is limited since the launch of attack. More resilient detection and defence models are typically more costly. We aim at predicting the likelihood of DDoS attacks by monitoring relevant text streams in social media, so that the level of defense can be adjusted dynamically for maximizing cost-effect. To our knowledge, this is a novel and challenge research question for DDoS rescue. Because the input of this task is a text stream rather than a document, information should be collected both on the textual content of individual posts. We propose a fine-grained hierarchical stream model to capture semantic information over infinitely long history, and reveal burstiness and trends. Empirical evaluation shows that social text streams are indeed informative for DDoS forecasting, and our proposed hierarchical model is more effective compared to strong baseline text stream models and discrete bag-of-words models.

There has been a recent line of work automatically learning scripts from unstructured texts, by modeling narrative event chains. While the dominant approach group events using event pair relations, LSTMs have been used to encode full chains of narrative events. The latter has the advantage of learning long-range temporal orders, yet the former is more adaptive to partial orders. We propose a neural model that leverages the advantages of both methods, by using LSTM hidden states as features for event pair modelling. A dynamic memory network is utilized to automatically induce weights on existing events for inferring a subsequent event. Standard evaluation shows that our method significantly outperforms both methods above, giving the best results reported so far.

In the multilingual World Wide Web, it is critical for Web applications, such as multilingual search engines and targeted international advertisements, to know what languages the user understands. However, online users are often unwilling to make the effort to explicitly provide this information. Additionally, language identification techniques struggle when a user does not use all the languages they know to directly interact with the applications. This work proposes a method of inferring the language(s) online users comprehend by analyzing their social profiles. It is mainly based on the intuition that a user’s experiences could imply what languages they know. This is nontrivial, however, as social profiles are usually incomplete, and the languages that are regionally related or similar in vocabulary may share common features; this makes the signals that help to infer language scarce and noisy. This work proposes a language and social relation-based factor graph model to address this problem. To overcome these challenges, it explores external resources to bring in more evidential signals, and exploits the dependency relations between languages as well as social relations between profiles in modeling the problem. Experiments in this work are conducted on a large-scale dataset. The results demonstrate the success of our proposed approach in language inference and show that the proposed framework outperforms several alternative methods.

Discourse relations may be either mononuclear or multi-nuclear. A mononuclear relation holds a nucleus and a satellite, the nucleus usually reflects the intention focus of the discourse, while the satellite represents supportive information. Although there are several previous works focused on analyzing discourse relations, only limited works focused on recognizing nuclearity between discourse units. In this paper, we propose a discourse unit nuclearity classifier in Chinese Discourse Treebank (CDTB). Our classifier considers the context of the two arguments, part of speech information, as well as word pair information as the features. The evaluations in CDTB corpus shows the effectiveness of our methodology and creates a baseline of 53.21% in accuracy.

Previous researches have placed emphasis on analyzing emotions in monolingual text, neglecting the fact that emotions are often found in bilingual or code-switching posts in social media. Traditional methods for the identification or classification of emotion fail to accommodate the code-switching content. To address this challenge, in this paper, we propose a multi-view learning framework to learn and detect the emotions through both monolingual and bilingual views. In particular, the monolingual views are extracted from the monolingual text separately, and the bilingual view is constructed with both monolingual and translated text collectively. Empirical studies demonstrate the effectiveness of our proposed approach in detecting emotions in code-switching texts.

Previous researches have focused on analyzing emotion through monolingual text, when in fact bilingual or code-switching posts are also common in social media. Despite the important implications of code-switching for emotion analysis, existing automatic emotion extraction methods fail to accommodate for the code-switching content. In this paper, we propose a general framework to construct and analyze the code-switching emotional posts in social media. We first propose an annotation scheme to identify the emotions associated with the languages expressing them in a Chinese-English code-switching corpus. We then make some observations and generate statistics from the corpus to analyze the linguistic phenomena of code-switching texts in social media. Finally, we propose a multiple-classifier-based automatic detection approach to detect emotion in the codeswitching corpus for evaluating the effectiveness of both Chinese and English texts.

Code-switching is commonly used in the free-form text environment, such as social media, and it is especially favored in emotion expressions. Emotions in codeswitching texts differ from monolingual texts in that they can be expressed in either monolingual or bilingual forms. In this paper, we first utilize two kinds of knowledge, i.e. bilingual and sentimental information to bridge the gap between different languages. Moreover, we use a term-document bipartite graph to incorporate both bilingual and sentimental information, and propose a label propagation based approach to learn and predict in the bipartite graph. Empirical studies demonstrate the effectiveness of our proposed approach in detecting emotion in code-switching texts.

Opinion summarization on conversations aims to generate a sentimental summary for a dialogue and is shown to be much more challenging than traditional topic-based summarization and general opinion summarization, due to its specific characteristics. In this study, we propose a graph-based framework to opinion summarization on conversations. In particular, a random walk model is proposed to globally rank the utterances in a conversation. The main advantage of our approach is its ability of integrating various kinds of important information, such as utterance length, opinion, and dialogue structure, into a graph to better represent the utterances in a conversation and the relationship among them. Besides, a global ranking algorithm is proposed to optimize the graph. Empirical evaluation on the Switchboard corpus demonstrates the effectiveness of our approach.

Cross-lingual sentiment classification aims to perform sentiment classification in a language (named as the target language) with the help of the resources from another language (named as the source language). Previous studies are prone to using all available data in the source language while using all data is observed to perform no better or even worse than using a partion of good data. In this paper, we propose a novel task called data quality controlling in the source language to select high quality samples from the source language. To tackle this task, we propose two kinds of data quality measurements: intra- and extra-quality measurements which are implemented with the certainty and similarity measurements respectively. The empirical studies demonstrate the effectiveness of the proposed approach to data quality controlling in the source language.

Sentiment classification is now a hot research issue in the community of natural language processing and the bag-of-words based machine learning approach is the state-of-the-art for this task. However, one important phenomenon, called polarity shifting, remains unsolved in the bag-of-words model, which sometimes makes the machine learning approach fails. In this study, we aim to perform sentiment classification with full consideration of the polarity shifting phenomenon. First, we extract some detection rules for detecting polarity shifting of sentimental words from a corpus which consists of polarity-shifted sentences. Then, we use the detection rules to detect the polarity-shifted words in the testing data. Third, a novel term counting-based classifier is designed by fully considering those polarity-shifted words. Evaluation shows that the novel term counting-based classifier significantly improves the performance of sentiment analysis across five domains. Furthermore, when this classifier is combined with a machine-learning based classifier, the combined classifier yields better performance than either of them.

Recently, sentiment classification has become a hot research topic in natural language processing. But most existing studies assume that the samples in the negative and positive categories are balanced, which might not be true in real applications. In this paper, we investigate sentiment classification tasks where the class distribution of the sam-ples is imbalanced. To handle the imbalanced problem, we propose a multi-strategy ensemble learning approach to this problem. Our ensemble approach integrates sample-ensemble, feature-ensemble, and classifier-ensemble by ex-ploiting multiple classification algorithms. Evaluation across four domains shows that our ensemble approach outper-forms many other popular approaches that handling imbal-anced classification problems, such as re-sampling and cost-sensitive approaches, and is proven effective for imbalanced sentiment classification.

Sentiment classification has undergone significant development in recent years. However, most existing studies assume the balance between negative and positive samples, which may not be true in reality. In this paper, we investigate imbalanced sentiment classification instead. In particular, a novel clustering-based stratified under-sampling framework and a centroid-directed smoothing strategy are proposed to address the imbalanced class and feature distribution problems respectively. Evaluation across different datasets shows the effectiveness of both the under-sampling framework and the smoothing strategy in handling the imbalanced problems in real sentiment classification applications.