Search

This paper reports how we build a Chinese Grammatical Error Diagnosis system in the NLPTEA-2018 CGED shared task. In 2018, we sent three runs with three different approaches. The first one is a pattern-based approach by frequent error pattern matching. The second one is a sequential labelling approach by conditional random fields (CRF). The third one is a rewriting approach by sequence to sequence (seq2seq) model. The three approaches have different properties that aim to optimize different performance metrics and the formal run results show the differences as we expected.

Customers are difficult to find and sometimes even more difficult to keep. If a company doesn’t pay attention to customer relationship management, it will spend lots of money to acquire them and then let them sleep. The goal of this study was to develop a sleeping customer detection system that collected users’ social network information and shopping behavior, and then classified them into 3 categories (sleeping customer, napping customer, and general customer). We collected the user information from January 01, 2015 to December 31, 2016. Support vector machine based classification was used. In this study, the overall accuracy was 81.7%. The results can help companies to reactivate sleeping customers as soon as possible.

With the prevalence of Social Networking Services (SNS), real-world consumption behaviors are influenced from reality to social networks. In order to utilize the information from social network, we need a concept space that can alter with application domain. In the presence of new vocabulary and continuously growing, and automatic ontology construction has been an important issue. There are previous studies concerning free-format ontology construction, enriching given ontologies from web or corpus sources, and construction of ontology from semi-structural corpora; among these studies, semi-structural corpora have been prevailing studies. In this thesis, we developed an adaptive framework for cross corpora on social collaborative editing, and we focus on semi-structural text mining in particular. The framework involves detection of named entity in a document, filtering of named entity, disambiguation detection, named entity expansion and ranking of the related named entity. We describe how this framework in detail and proposed method for each stage, and the metrics in the previous studies and the one we used for evaluation. We then discuss the evolution and quality of concept space. Our proposed framework made real-world corpora computationally possible, and a dynamic concept space is generated from this framework. It could deal with more diverse domains and languages, and for pragmatic real-world applications, our method shows better flexibility than previous studies. The research result can avoid the costly mistake of designing various mechanism and provide user adoption analysis and digital marketing strategy via online/offline in adopting this kind of innovation service.

The increasing popularity of social networking services and the mobile devices have changed people’s life and brought the new business challenges. The goal of this study was to analyze the characteristics and purchase probability of different customers groups in the Shimen shopping district and extract business value. We extracted a point-earning app’s user records about the Shimen shopping district. The dates of those user records were between June 11, 2015 and July 12, 2015. Furthermore, we collected the Facebook information of the users. All statistical procedures were performed with our Customer Behavior Analysis System. The main customers in Shimen shopping district were younger groups. Different groups of users had divergent favorites because of their age and gender. The popular trend nowadays in the younger people and the reward points the shop gave may affect the conversion rate of a shop. Many people shopped at several similar stores in one day and that might mean they shopped purposefully not blindly. After customizing the point-earning app’s assignments and promotions by the analysis results, we got an up to 6% conversion rate improvement.

The objective of the study is to generate the product hierarchical categories in e-commerce, particularly for e-commerce giants such as Taobao or Jingdong. For e-commerce websites the amount of products is huge, and a hierarchical structure is necessary for consumers to browse them. We find that there are two problems in the current websites: firstly, the hierarchy is shallow; there are often too many products in the same category, it is hard for a consumer browse them. Secondly, the hierarchy is constructed manually, when new products come, it is hard to update the hierarchy. Based on the product description analysis, it is possible to solve the problems. In this study, we will use the deep learning word embedding technology and clustering algorithm to construct a deeper product hierarchy automatically. The results will help the customers to choose products with a more clear structure and also help the e-commerce company to save the maintaining effort on the product hierarchy.

We introduce a method for learning to predict text and grammatical construction in a computer-assisted translation and writing framework. In our approach, predictions are offered on the fly to help the user make appropriate lexical and grammar choices during the translation of a source text, thus improving translation quality and productivity. The method involves automatically generating general-to-specific word usage summaries (i.e., writing suggestion module), and automatically learning high-confidence word- or phrase-level translation equivalents (i.e., translation suggestion module). At runtime, the source text and its translation prefix entered by the user are broken down into n-grams to generate grammar and translation predictions, which are further combined and ranked via translation and language models. These ranked prediction candidates are iteratively and interactively displayed to the user in a pop-up menu as translation or writing hints. We present a prototype writing assistant, TransAhead , that applies the method to a human-computer collaborative environment. Automatic and human evaluations show that novice translators or language learners substantially benefit from our system in terms of translation performance (i.e., translation accuracy and productivity) and language learning (i.e., collocation usage and grammar). In general, our methodology of inline grammar and text predictions or suggestions has great potential in the field of computer-assisted translation, writing, or even language learning.

In recent years, due to the rapid development of e-commerce, personalized recommendation systems have prevailed in product marketing. However, recommendation systems rely heavily on big data, creating a difficult situation for businesses at initial stages of development. We design several methods — including a traditional classifier, heuristic scoring, and machine learning — to build a recommendation system and integrate content-based collaborative filtering for a hybrid recommendation system using Co-Clustering with Augmented Matrices (CCAM). The source, which include users' persona from action taken in the app & Facebook as well as product information derived from the web. For this particular app, more than 50% users have clicks less than 10 times in 1.5 year leading to insufficient data. Thus, we face the challenge of a cold-start problem in analyzing user information. In order to obtain sufficient purchasing records, we analyzed frequent users and used web crawlers to enhance our item-based data, resulting in F-scores from 0.756 to 0.802. Heuristic scoring greatly enhances the efficiency of our recommendation system.

In this paper we report how we improve our Chinese grammar error detection system. We focus on the recognition of word selection error, which is the hardest error type defined in 2015 NLP-TEA shared task CGED. Four major error types was defined in CGED shared task, including redundant word error, missing word error, word selection error and word disorder error. Based on the conditional random fields (CRF) model, our system trains a linear tagger that can be used to recognize the errors in learners' essays. In 2015 CGED shared task, we have found that our system can achieve high precision and accuracy. Our previous system used the features in the sentence only, however, it is not sufficient to recognize the word selection error. In this paper, we propose that by integrating the collocation feature into the CRF model, our system can improve the performance on the recognition of the selection errors.

Various writing assistance tools have been developed through efforts in the areas of natural language processing with different degrees of success of curriculum integration depending on their functional rigor and pedagogical designs. In this paper, we developed a system, WriteAhead, that provides six types of suggestions when non-native graduate students of English from different disciplines are composing journal abstracts, and assessed its effectiveness. The method involved automatically building domain-specific corpora of abstracts from the Web via domain names and related keywords as query expansions, and automatically extracting vocabulary and n-grams from the corpora in order to offer writing suggestions. At runtime, learners' input in the writing area of the system actively triggered a set of corresponding writing suggestions. This abstract writing assistant system facilitates interactions between learners and the system for writing abstracts in an effective and contextualized way, by providing suggestions such as collocations or transitional words. For assessment of WriteAhead, we compared the writing performance of two groups of students with or without using the system, and adopted student perception data. Findings show that the experiment group wrote better, and most students were satisfied with the system concerning most suggestion types, as they can effectively compose quality abstracts through provided language supports from WriteAhead.

We introduce a method for learning to translate out-of-vocabulary (OOV) words. The method focuses on combining sublexical/constituent translations of an OOV to generate its translation candidates. In our approach, wildcard searches are formulated based on our OOV analysis, aimed at maximizing the probability of retrieving OOVs’ sublexical translations from existing resources of Machine Translation (MT) systems. At run-time, translation candidates of the unknown words are generated from their suitable sublexical translations and ranked based on monolingual and bilingual information. We have incorporated the OOV model into a state-of-the-art machine translation system and experimental results show that our model indeed helps to ease the impact of OOVs on translation quality, especially for sentences containing more OOVs (significant improvement).

The goal of this study was to analyze the characteristics and purchase probability of different customers groups in the retail store POYA. We abstracted CheckMe app user records about the retail store POYA. The dates of those user records were between January 2015 and April 2015. Furthermore, we collected the Facebook information of the users. All statistical procedures were performed with our Persona Analysis Platform. The purchase probability of female subjects was nearly twice than male subjects in POYA. Although, the main subjects were 18–39 years old, the purchase probability of 40–44 years old subjects was higher than others. Most of the customers were from the metropolises. However, the purchase probability of the subjects in Hsinchu and Pingtung was higher than that in other cities. After we provided analysis results to POYA, it improved its promotions and got an up to 10 % conversion rate improvement.

In the age of information network explosion, users can be connected to various forms of social networking sites whenever and wherever through the Internet, And through the social networking platform to interact with people. Through the media in this way, people know each other's interests and preferences by the relationship between social friends, but for many international companies, it is important to know how to understand individual preferences, because when you know the individual preference information, it can carry out personal preference for advertising, product recommendation, article recommended and other diversified personal social service, which can increase the click rate and exposure of the products, better close to the needs of the user's life. Therefore, with the above through social science and technology development trend arising from current social phenomenon, research of this thesis, mainly expectations for analysis by the information of interaction between people on the social network, such as: user clicked fan page, user's graffiti wall message information, friend clicked fan page etc. three kinds of personal information for personal preference analysis, and from this huge amount of personal data to find out corresponding diverse group for personal preference category. We can by personal preference information for diversify personal advertising, product recommendation and other services. The thesis at last through the actual business verification, the research can improve website browsing pages growth 11%, time on site growth 15%, site bounce rate dropped 13.8%, product click through rate growth 43%, more fully represents the results of this research fit the use's preference.

We introduce a method that extracts keywords in a language with the help of the other. The method involves estimating preferences for topical keywords and fusing language-specific word statistics. At run-time, we transform parallel articles into word graphs, build crosslingual edges for word statistics integration, and exploit PageRank with word keyness information for keyword extraction. We apply our method to keyword analysis and language learning. Evaluation shows that keyword extraction benefits from cross-language information and language learners benefit from our keywords in reading comprehension test.

This paper reports how to build a Chinese Grammatical Error Diagnosis system based on the conditional random fields (CRF). The system can find four types of grammatical errors in learners’ essays. The four types or errors are redundant words, missing words, bad word selection, and disorder words. Our system presents the best false positive rate in 2015 NLP-TEA-2 CGED shared task, and also the best precision rate in three diagnosis levels.

We introduce a method that extracts keywords in a language with the help of the other. In our approach, we bridge and fuse conventionally irrelevant word statistics in languages. The method involves estimating preferences for keywords w.r.t. domain topics and generating cross-lingual bridges for word statistics integration. At run-time, we transform parallel articles into word graphs, build cross-lingual edges, and exploit PageRank with word keyness information for keyword extraction. We present the system, BiKEA , that applies the method to keyword analysis. Experiments show that keyword extraction benefits from PageRank, globally learned keyword preferences, and cross-lingual word statistics interaction which respects language diversity.

Social media has become the most convenient space to retrieve the tremendous consumers' experience, opinion and preference–toward each brands, products or even specific features. The real-time and big amount characteristics of Social media provide great opportunity for producer to know their customers (and potential ones) well. This paper proposes an Online to Offline (O2O) Commerce Service Model and takes the social relationship dashboard as an pilot study, which can help retailers or brands to understand their customers via social network existing data (especially Facebook for this case) by which we can adapt the current social commerce marketing strategy more quickly and responsively.

The hand-held systems and the wireless Internet access are widely available in recent years. However, the mobile learning with web content is still inconvenient. For example, the information has not well organized and it is uneasy to surf on the small screen of handheld device. We propose a mobile system based on the content of Wikipedia. Wikipedia is a free content resource and has abundant contents of text and pictures. We use the Wikipedia wrapper that we developed before, to develop the mobile-learning interface of cross-language and cross-platform applications. Our system can present the content of Wikipedia on the small screens of PDA, and can be use on mobile learning. A teaching scenario of mobile learning during a museum visiting is discussed in this paper.

In this research, we propose a Chinese essay error detection system that can be used in online writing tutorial environment. The system consists of word segmentation, template module, and language model module. We build the system with a dictionary for word segmentation and generation of detection templates from news corpus. We also gather the character error probabilities from students’ essays to build error model. Error types include pronunciation-related errors and composition-related errors. Our system provides two operating modes for users with different goals. For example, we can help students learn to write effectively by providing high precision correction mode; for teachers, we provide high correction detection mode which can help teachers to check students’ essays.