Automatic recognition of customer complaints on products or services that they purchase can be crucial for the organizations, multinationals and online retailers since they can exploit this information to fulfil their customers’ expectations including managing and resolving the complaints. Recently, researchers have applied supervised learning strategies to automatically identify users’ complaints expressed in English on Twitter. The downside of these approaches is that they require labeled training data for learning, which is expensive to create. This poses a barrier for them being applied to low-resource languages and domains for which task-specific data is not available. Machine translation (MT) can be used as an alternative to the tools that require such task-specific data. In this work, we use state-of-the-art neural MT (NMT) models for translating Hindi reviews into English and investigate performance of the downstream classification task (complaints identification) on their English translations.
In this paper, we examine and analyze the challenges associated with developing and introducing language technologies to low-resource language communities. While doing so, we bring to light the successes and failures of past work in this area, challenges being faced in doing so, and what they have achieved. Throughout this paper, we take a problem-facing approach and describe essential factors which the success of such technologies hinges upon. We present the various aspects in a manner which clarify and lay out the different tasks involved, which can aid organizations looking to make an impact in this area. We take the example of Gondi, an extremely-low resource Indian language, to reinforce and complement our discussion.
We present a quantitative, data-driven machine learning approach to mitigate the problem of unpredictability of Computer Science Graduate School Admissions. In this paper, we discuss the possibility of a system which may help prospective applicants evaluate their Statement of Purpose (SOP) based on our system output. We, then, identify feature sets which can be used to train a predictive model. We train a model over fifty manually verified SOPs for which it uses an SVM classifier and achieves the highest accuracy of 92% with 10-fold cross-validation. We also perform experiments to establish that Word Embedding based features and Document Similarity-based features outperform other identified feature combinations. We plan to deploy our application as a web service and release it as a FOSS service.
We investigate the problem of predicting the competence of users of the crowdsourcing platform Zooniverse by analyzing their chat texts. Zooniverse is an online platform where objects of different ...
This paper deals with semantic enrichment of textual resources by means of automatically generated named entity recognizers-linkers and advanced indexing and searching mechanisms that can be integrated into various information retrieval and information extraction systems. It introduces a new system transforming Wikipedia and other available sources into task-specific knowledge bases and employs contextual information to build state-of-the-art entity disambiguation components. Although some components are language-dependent (for example, that responsible for the morphology analysis or the semantic role identification), they can be easily replaced by existing tools providing specific functions. As a case study, we demonstrate an instantiation of the sys-tem for the task of semantic annotation of Czech bibliographic databases in the context of the CPK project. We particularly stress the role of problem-specific knowledge sources that can be easily integrated into our system and play a key role in the success of the tool in real applications.
A system, Magi, is proposed, which analyses simple addition/subtraction arithmetic word problems expressed in English, represents them in the form of schemas and sets, reasons with set cardinalities and presents the final answer in English phrases. It also provides simple explanations. This work presents a study of the features of a knowledge-based system used for solving such a task. It has been evaluated and has been found to perform better than current knowledge-based systems for similar problems.
Cyber-bullying may be defined as the employment of technological means for the purpose of harassing, threatening, embarrassing, or targeting a particular person. It is also possible for Cyber-bullying to have occurred accidentally. One of the major challenges in identifying cyber-bullying or cyber-aggressive comments is to detect a sender’s tone in a particular text message, email or comments on social media, since what a person may consider to be a joke, may act as a hurting insult to another. Nevertheless, cyber-bullying may prove to be non-accidental in specific cases where a repetition in the pattern of text in emails, messages, and online posts is existent. In order to curb such a social threat, this Paper proposes the usage of a combination of document embeddings along with different supervised machine learning algorithms to get optimized results in flagging cyber-aggressive comments. Extensive experimentation indicates that the SVM model with rbf kernel combined with document embeddings is capable of efficiently classifying unseen test comments with an accuracy score of 88.465 % and has surpassed other models in various evaluation metrics.
Morphological disambiguation, particularly for morphologically rich languages, is a crucial step in many NLP tasks. Morphological analyzers provide multiple analyses of a word, only one of which is true in context. We present a language-agnostic deep neural system for morphological disambiguation, with experiments on Hindi. We achieve accuracies of around 95.22% without the use of any language-specific features or heuristics, which outperforms the existing state of the art. One contribution through this work is building the first morphological disambiguation system for Hindi. We also show that using phonological features can improve performance. On using phonological features and pre-trained word vectors, we report an accuracy of
This paper sets out to investigate ways of normalizing noisy text that appear on social media platforms like Facebook, Twitter, Whatsapp, etc. We proposed a deep learning based approach to text normalization using Recurrent Neural Network (RNN) based Encoder–Decoder architecture with Long Short Term Memory (LSTM). To circumvent the unavailability of suitable large noisy–clean parallel dataset, we developed synthetic datasets. We trained and evaluated the proposed model on our synthetic datasets and the WNUT1 shared task dataset. The uniqueness of our approach is in the use of synthetic datasets in a transfer learning approach for improving the performance of text normalization based on deep neural models. Our transfer learning based deep neural model produced state-of-theart results (F1 score 0.9098) outperforming the previous best performing model on the WNUT test set by 7%.
This paper presents a robust system for the automatic extraction of bio-molecular events from scientific texts. Event extraction provides information in the understanding of physiological and pathogenesis mechanisms. Event extraction from biomedical literature has a broad range of applications, such as knowledge base creation, knowledge discovery. Automatic event extraction is a challenging task due to ambiguity and diversity of natural language and linguistic phenomena, such as negations, anaphora and coreferencing leading to incorrect interpretation. In this work a machine learning based approach has been used for the event extraction. The methodology framework proposed in this work is derived from the perspective of natural language processing. The system includes a robust anaphora and coreference resolution module, developed as part of this work. An overall F-score of 54.25% is obtained, which is an improvement of 4% in comparison with the state of the art systems.
