anishacharya/nlp-cs388-ut

Common Libraries developed in "PyTorch" for different NLP tasks. Sentiment Analysis, NER, LSTM-CRF, CRF, Semantic Parsing

NLP-CS388-UT

Please cite this repository if you use any of the code in your research

This repository is a demonstration on how to modularly write re-usable code for NLP for research/production. Kind of trying to bridge/merge my experiences in both academia and industry. A single codebase with many shared functions across several common NLP tasks.

I hope you benefit from this to quickly spin up new projects and not write everything from scratch. Also please NEVER write flat python main.py files with 2000 lines of code ( A huge problem I have noticed in academia and also among many researchers in industry) .

Remember : Well writen code = Highly productive research

Framework PyTorch

Additonally this contains Four Project Reports.
http://www.cs.utexas.edu/~gdurrett/courses/fa2019/cs388.shtml.

Feature Engineered Named Entity Recognition

Named Entity Recognition(NER) is a fundamen-tal NLP task where the objective is to identify thenamed entities in a piece of text. In this work,we focus on designing powerful features for NER,which leads to reasonably good accuracy withoutthe need of a complex model. We tried our ap-proach on CONLL-2003 NER dataset which hasfour class of named entities: person, organiza-tion, location, and miscellaneous. We just focuson identifying instances of the person label in iso-lation for this work.

Named Entity Recognition as a sequence labeling problem

Named Entity Recognition(NER) is a fundamen-tal NLP task where the objective is to identify thenamed entities in a piece of text. In this work, wefocus on designing powerful features and modelsfor NER, which leads to reasonably good accu-racy without the need of a complex model. Wetried our approach on CONLL-2003 NER datasetwhich has four class of named entities: person, or-ganization, location, and miscellaneous.

Deep Ordered and Unordered Syntactic Compositionfor Sentiment Analysis

With the success of Deep Learning NLP re-search has moved from shallow-models on sparse-feature-space to deep-models on dense word-embedding-space. In this work we explore twopopular deep learning approaches on a sentimentclassification task. We evaluate our models onRotten Tomatoes movie reviews dataset, which as-signs a binary label to each movie review.

Semantic Parsing as a Seq2Seq Problem

Semantic parsing is the task of translating textto a formal meaning representation such as log-ical forms or structured queries. There are sev-eral decades of history associated with the classi-cal NLP task of semantic parsing.In this project we pose the problem of semanticparsing in a machine tranlation framework wherethe source is the standard text input while the tar-get being the logical form. An intuitive way tothink about it to think of logical form as anotherlanguage and then treat this as a translation prob-lem.In this setting, we use a sequence to sequence styleencoder-decoder architecture to do semnatic pars-ing. We particularly work on the geo-query datasetwhich in the downstream is used to do QA. Theinputs are plain english form questions while theoutputs/targets are logical forms that can be usedto query knowledge-graphs to get answers.