Natural Language Processing through ML : CSE Department Webinar
The Department of Computer Science and Engineering, NIT Rourkela organised a Seminar on leveraging Natural Language Processing through Machine Learning from 15th -17th October 2020 under the guidance of Prof Bidyut Kumar Patra and PhD scholars Sumanto Dutta and Sudhansu Bala Das. Ministry of Electronics and Information Technology (MeiTy), Government of India sponsored this conference. The department has been equipped with modern laboratories such as DataMining, Image Processing & Cluster Computing, Advanced Database Engineering labs, Data Communication, Information Security, RecIFand Software Engineering for pursuing research in view of technological advancement. The 3-day long webinar witnessed numerous prominent professors, academicians and researchers in Artificial Intelligence and Machine Learning from across India as its speakers. The event provided students as well as professionals with a golden opportunity to expand their knowledge in this domain.
Natural Language Processing (NLP) plays a vital role to address the various challenges related to understanding the human language. NLP has various tasks like text summarization, named entity recognition, relationship extraction, language translation and sentiment analysis (SA) and much more. This conference aimed to deliver a holistic view of maximizing usage of NLP in our day to day life in many forms. The take away from this conference was an in-depth understanding of NLP to solve problems by machines which are helpful for our daily life.
Three sessions were organised every day with an average duration of 1 hour and 30 minutes. The conference was open to all and the participants were provided with the E-Certificate.
Topic- Word embedding
The first session of the inaugural day was addressed by Dr Sanasam Ranbir Singh from IIT Guwahati. The session was on the basics of Word Embedding in the field of Machine Learning. He started with explaining the use of word embedding in text analysis such as Text Recommendations, classification and Image/video tagging. He vividly explained the traditional form of text representation in Machine Learning and how words and documents are represented in the form of vectors, used in matrices to classify embedded documents. He also discussed the problems with the document term representations. He further discussed Principal Component Analysis (PCA), Eigenvalue decomposition for square matrix and Singular value decomposition for rectangular value matrix. Prof. Singh also gave a brief introduction to Neural Networking followed by working of Word2vec which is a technique for natural language processing. In the end, he showed the pre-loaded google word2vec and gave an example of how it works.
After the first session of day 1, Prof Animesh Biswas, Director, NIT Rourkela addressed and welcomed the enthusiastic minds from all over the country, who participated in the seminar to learn and share the experience and work together in these three days. He congratulated the Department of Computer Science for taking initiative to conduct such seminars on such demanding topics in the pandemic situation and wished everyone luck. It was followed by a vote of thanks by Prof Sathya Babu and then proceeding to the next session.
Topic- Language Analysis
This session, addressed by Prof. Pawan Goyal, Associate Professor, IIT Kharagpur, was primarily on the issue of ‘Graph-based framework for Structured Prediction Tasks in Sanskrit’. Prof. Goyal, with many distinguished achievements in the field of Artificial Intelligence, talked about the significance of Sanskrit language in Indian culture, and its grammatical tradition. He underlined the use of Machine Learning in Sanskrit computational linguistics. He explained that the readers, though want to read the language, are not able to clearly understand it. So, making a computer interface to resolve the various arduous tasks of word segmentation, morphological analysis, poetry to prose conversion etc becomes imperative. The challenges involve free order word nature, very little task-specific annotated data, compound words, among others.
Prof. Goyal elucidated on basic processing for NLP, starting from word tokenization to processing texts. He also explained ‘Sandhi-Phonetic transformation at word boundaries’. With that, he gave solution models, which are computationally efficient too.
Topic - Scientific Literature Processing
This session was addressed by Prof Tanmoy Chakraborty from IIIT Delhi. He vividly described scientific documents (paper) mining. He explained about the citation information and its uses by the authors of different books. Scientists use references in much the same way, drawing on available information to conduct research. When scientists present their new ideas and results to the community, they are expected to support their ideas with knowledge of the scientific literature and the work that has come before them.
Prof Chakraborty mentioned that scientific literature is of central importance to the growth and development of science as a whole.
Topic- Prosody Modelling for Speech Synthesis
This session was addressed by Prof. Dr S. K. Das Mandal from IIT Kharagpur, and an achieved genius in the field of Speech Technology and Signal Processing.
Prof. Das mostly covered the application and research prospects of the arena. He explained the importance of speech and text in this age of information society. He vividly made the attendees understand the major tasks for the machine to analyse the speech and convert it into appropriate formats. He also gave information on segmental and suprasegmental features of speech.
The major highlight was the prosody disambiguation and its role in speech analysis and elucidated it with examples and differences between various aspects of it and the code system. He also told about prosodic structure and prosody modelling. The chronological features of it were also explained.
He elucidated on various research prospects of the field.
Topic- Multi-Modal Dialogue management system
The second session of the day witnessed an interesting lecture by Dr.Sriparna Saha, IIT Patna on Multimodal Dialogue management system which basically explains the working of a chatbot. She covers the basic working and components of a Chatbot through the seminar.
Professor started with the basic definition of dialogue, basics of how the Dialogue system works, covering the Dialogue act classifications where she explains on identifying the type of dialogue by not only using voice but also facial expressions. Further giving a vivid description of the Hierarchical Dialogue Policy learning and how emotion data (EMOTyDA) are used. Prof. further covers the Dialogue management strategy for multi-domain, multi-intent conversations and how Sentiment aware multi-modal dialogue policy works using Hierarchical Reinforcement Learning. The session was concluded with a Q&A session from students which were answered by the professor descriptively and lucidly.
Topic: Machine Translation
The day was followed with an interesting session by Prof Partha Pakray, NIT Silchar on Machine Translation. He started with an introduction and challenges in Machine Translation. Professor further covered the Machine Translation tools, current trends and future directions on the same. He mentions that Machine translation is challenging given the inherent ambiguity and flexibility of human language. Statistical machine translation replaces classical rule-based systems with models that learn to translate from examples whereas Neural machine translation models fit a single model rather than a pipeline of fine-tuned models and currently achieve state-of-the-art results.
He concluded the session by sharing the references he has used so that the students can research more on the topic.
Topic: Deep Learning for Sentiment Analysis
The foremost lecture of day 3 was an insightful session by Dr Asif Ekbal, Associate Professor (CSE) from IIT Patna on the topic Deep Learning for Sentiment Analysis. Starting with a brief introduction and background of sentiment analysis, he explained the more recent and fine-grained sentiment analysis method known as Aspect Based Sentiment Analysis (ABSA) with examples and differences from the previous method. Going on he explained the different Deep learning methods to approach sentiment analysis problems like Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long short term memory, Attention Mechanism and memory learning and how those are better than the traditional Machine Learning methods. He further explained the Inter aspect Dependency of Sentiment Analysis, target specific representation, Context-dependent target and Multilinguality. He also discussed Multimodal Sentiment Analysis. The session concluded with a Q & A session.
Topic: Text Summarization
The day was followed by Prof Sriparna Saha, Associate Professor and Associate Dean Research and Development in Department of Computer Science and Engineering in IIT Patna. She explained how summaries reduce reading time in researching documents. Using automatic or semi-automatic summarization systems enables commercial abstract services to increase the number of text documents they are able to process. Insights related to Single-objective optimization (SOO) and Multi-objective optimization (MOO) were also given. The session also had various real-life problem discussions which led the participants to ponder more on the topic. The session concluded with a Q & A session.
Topic: Hands-on Machine Translation
The last session of the seminar was an extension of the third session of Day 2 by Prof Partha Pakray, NIT Silchar, where he vividly demonstrated the hands-on experience on machine translation, and it was basically an extension of his previous session.
With this, the three-days long, insightful webinar on very brilliant topics in the field of computer science, came to a conclusion.
Words from the faculty coordinator, Prof. Bidyut Kumar Patra :
This three-day seminar aims to deliver a holistic view of maximizing usage of natural language processing (NLP) in our day to day life in many forms. The take away from this seminar is an in-depth understanding of different NLP problems in our day to day life. Ministry of Electronics and Information Technology (MeiTy), Government of India sponsored this conference. Seven expert speakers from various IITs and NITs accepted our invitation and spent their valuable time with us explaining different topics of NLP. On day 1, speakers spoke about the representation and analysis of language with applicability in scientific language processing. Whereas in day 2, speakers talked about prosody modelling for speech synthesis, multimodal dialogue system and machine translation. On last day, speakers presented about sentiment analysis, text summarization with a hands-on session on machine translation. Total of 163 participants registered for these sessions, where 113 participants are from various institutes all over the country and rest are from our institute. Daily 70-80 people attended the live session. I earned colossal experience in conducting this seminar and realized that these type of seminars are needed for our students in the near future!
The Seminar received appreciation and positive feedback from all over. This is what one of the students named Dibyajyoti Jena from Stewart Science College, Cuttack had to say for the seminar:
Undoubtedly all the sessions of were wonderful. Thank you so much Sir for giving us this wonderful opportunity to learn and explore many new things from eminent speakers of reputed institutions just by sitting at home. Looking for for more such events in the near future.
Team Monday Morning compliments Prof Bidyut Kumar Patra and his PhD scholars for successfully conducting this webinar and wishes them the best for all future endeavours.