In this project, we aim at building a system that can automatically generate product reviews in a domain, such as movie reviews or book reviews. Two different approaches will be used. One of the approaches is a pattern-based approach, where the reviews are formed by filling in the slots of predefined patterns. The second one is an encoder-decoder approach, where the reviews will be generated by a deep learning model [1]. In the first part of the project, the literature on automatic review generation will be surveyed.
In this project, we aim at building a comprehensive word embedding repository for the Turkish language. Using each of the state-of-the-art word embedding methods, embeddings of all the words in the language will be formed using a corpus. First, the three commonly-used embedding methods (Word2Vec, Glove, Fasttext) will be used and an embedding dictionary for each one will be formed. Then we will continue with context-dependent embedding methods such as BERT and Elmo. Each method will be applied with varying parameters such as different corpora and different embedding dimensions.
Named entity recognition aims to detect entities that refer to people, locations, organizations and similar in a given sentence.
This project involves reimplementing a recent NER tagger that is shown to surpass the state-of-the-art performance for morphologically rich languages [1].
We will employ a software framework that is specific to NLP to easily build, train, evaluate and deploy the new tagger, i.e. Stanza, Flair or Huggingface.
We will also add some new features to exploit all types of word embeddings easily.
Edge systems can be thought of as micro-cloud infrastructures that serve devices in proximity. Devices with insufficient computational capacities (AR/VR glasses, mobile gadgets, smartphones depending on the application) can augment their compute power using these edge servers.
Keyword extraction is the process of automatically identifying the important words or phrases in a given text. In this project, you will design and implement a keyword extraction system using deep learning models. You will start by replicating the system described in the paper “Deep Keyphrase Generation” using a recurrent neural network model. Then you will continue with alternative deep learning models. Finally, the pros and cons of these architectures will be compared.
