Journal Cover
Studia Universitatis Babeș-Bolyai Informatica
Number of Followers: 0  

  This is an Open Access Journal Open Access journal
ISSN (Print) 1224-869X - ISSN (Online) 2065-9601
Published by Babes-Bolyai University Homepage  [4 journals]
  • Can You Guess the Title' Generating Emoji Sequences for Movies

    • Authors: A Bajcsi, B Botos, P Bajko, Z Bodo
      Pages: 5 - 20
      Abstract: In the culture of the present emojis play an important role in written/typed communication, having a primary role of supplementing the words with emotional cues. While in different cultures emojis can be interpreted and thus used differently, a small set of emojis have clear meaning and strong sentiment polarity. In this work we study how to map natural language texts to emoji sequences, more precisely, we automatically assign emojis to movie subtitles/scripts. The pipeline of the proposed method is as follows: first the most relevant words are extracted from the movie subtitle, and then these are mapped to emojis. In order to perform the mapping, three methods are proposed: a lexical matching-based, a word embedding-based and a combined approach. To demonstrate the viability of the approach, we list some of the generated emojis for a randomly selected movie subset, showing also the deficiencies of the method in generating guessable sequences. Evaluation is performed via quizzes completed by human participants.
      PubDate: 2022-07-03
      DOI: 10.24193/subbi.2022.1.01
      Issue No: Vol. 67, No. 1 (2022)
  • Multiple Types of AI and Their Performance in Video Games

    • Authors: I Prajescu, A.-D. Calin
      Pages: 21 - 36
      Abstract: In this article, we present a comparative study of Artificial Intelligence training methods, in the context of a racing video game. The algorithms Proximal Policy Policy Optimization (PPO), Generative Adversarial Imitation Learning (GAIL) and Behavioral Cloning (BC), present in the Machine Learning Agents (ML-Agents) toolkit have been used in several scenarios. We measured their learning capability and performance in terms of speed, correct level traversal, number of training steps required and we explored ways to improve their performance. These algorithms prove to be suitable for racing games and the toolkit is highly accessible within the ML-Agents toolkit.
      PubDate: 2022-07-03
      DOI: 10.24193/subbi.2022.1.02
      Issue No: Vol. 67, No. 1 (2022)
  • Romanian Question Answering Using Transformer Based Neural Networks

    • Authors: B.-A. Diaconu, B Lazar-Lorincz
      Pages: 37 - 44
      Abstract: Question answering is the task of predicting answers for questions based on a context paragraph. It has become especially important, as the large amounts of textual data available online requires not only gathering information but also the task of findings specific answers to specific questions. In this work, we present experiments evaluated on the XQuAD-ro question answering dataset that has been recently published based on the translation of the SQuAD dataset into Romanian. Our bestperforming model, Romanian fine-tuned BERT, achieves an F1 score of 0.80 and an EM score of 0.73. We show that fine-tuning the model with the addition of the Romanian translation slightly increases the evaluation metrics.
      PubDate: 2022-07-03
      DOI: 10.24193/subbi.2022.1.03
      Issue No: Vol. 67, No. 1 (2022)
  • Music Recommendations Based on User's Mood Using Convolutional Neural

    • Authors: A Petrescu
      Pages: 45 - 60
      Abstract: This paper proposes a method for music recommendations using emotions, using deep learning techniques. The method is composed of two modules. The emotion detection module, which utilizes a hybrid architecture involving a Convolutional Neural Network (CNN) and a Reccurent Neural Network using Long-Short Term Memory (LSTM) Cells. We compared individual architectures of CNNs and LSTMs against our hybrid approach, outperforming them during experiments. We evaluated the modules on our own data set, created using Spotify’s API and containing 2028 songs from different genres and linguistic families, labeled with valence and arousal values. The model also outperforms other related approaches, however we did not evaluate them on the same data set. The predictions are used by the second module, for which we proposed a simple method of ordering the results based on the similarity to user’s input.
      PubDate: 2022-07-03
      DOI: 10.24193/subbi.2022.1.04
      Issue No: Vol. 67, No. 1 (2022)
  • A Dynamic Approach for Railway Semantic Segmentation

    • Authors: A.-R. Alexandrescu, A. Manole
      Pages: 61 - 76
      Abstract: Railway semantic segmentation is the task of highlighting rail blades in images taken from the ego-view of the train. Solving this task allows for further image processing on the rails, which can be used for more complex problems such as switch or fault detection. In this paper we approach the railway semantic segmentation using two deep architectures from the U-Net family, U-Net and ResUNet++, using the most comprehensive dataset available at the time of writing from the railway scene, namely RailSem19. We also investigate the effects of image augmentations and different training dataset sizes, as well as the performance of the models on dark images. We have compared our solution to other approaches and obtained competitive results with larger scores.
      PubDate: 2022-10-03
      DOI: 10.24193/subbi.2022.1.05
      Issue No: Vol. 67, No. 1 (2022)
School of Mathematical and Computer Sciences
Heriot-Watt University
Edinburgh, EH14 4AS, UK
Tel: +00 44 (0)131 4513762

Your IP address:
Home (Search)
About JournalTOCs
News (blog, publications)
JournalTOCs on Twitter   JournalTOCs on Facebook

JournalTOCs © 2009-