Design and Architectures of Data-Centric and Knowledge Based Systems

Permanent URI for this collectionhttps://hdl.handle.net/10125/107531

Browse

Recent Submissions

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item type: Item ,
    Cheshire: A New Believable Chat Bot Using AIML, LSA, Emotions from Personalities, and Voice Recognition and Synthesizer
    (2024-01-03) Suh, Sang; Baxter, Stephen; Carbone, John; Then, Patrick
    Since the 1950s, people have been trying to create a more believable chatbot. The Standard Turing Test (STT) has generally been used to test them. Development of chatbot initiated with pattern recognition with Eliza in 1966 and PARRY in 1972, further with AI by Jabberwacky, and AIML with ALICE in 1995. Since then, people have tried adding nonverbal features, personalities, and audio input and output features. The goal of this research is to use these advancements to create a chatbot believable enough to pass the STT. To do this in a different way than most other chatbots, this new chatbot will use AIML with LSA to generate a response, derive and use the emotional ton of the user input along with a selected personality to apply an emotional ton to the response, and provide a means for the user to talk to the chatbot and for the chatbot to talk back.
  • Thumbnail Image
    Item type: Item ,
    Sentimental Analysis of Movie Tweet Reviews Using Machine Learning Algorithms
    (2024-01-03) Kari, Hemanth Kumar
    Sentiment analysis stands as a prominent tool within microblogging platforms, gaining substantial traction as a means to discern public opinion and sentiment across various topics, including movie tweet reviews. In response to this demand, the study introduces a robust system architecture that incorporates an array of algorithms, ranging from Multinomial Naive Bayes and Support Vector Machine (SVM) to K-Nearest Neighbors (KNN), Bernoulli’s Naive Bayes, and Random Forest. This architecture is meticulously trained using annotated Twitter data, methodically excluding non-opinionated content while precisely identifying sentiment. Thorough experimentation underscores the effectiveness of our methodology. To accomplish this, we curate an extensive data set of movie-related tweets, each carefully labeled with sentiments spanning positive, negative, or neutral tones. The methodological framework involves intricate text preprocessing steps, encompassing tokenization, stemming, and the removal of extraneous stop words. This facilitates the extraction of essential features and the conversion of raw text into numerical representations suitable for machine learning. Our sentiment classification modeling employs a diverse ensemble of machine learning algorithms, including Naive Bayes, Support Vector Machines, and Recurrent Neural Networks. The assessment involves a range of metrics such as accuracy, precision, recall, and F1-score, supported by rigorous techniques like cross-validation to enhance the dependability and robustness of results. Our unique contribution lies in the strategic deployment of algorithms and a resilient system architecture adept at surmounting the challenges inherent to microblogs. We emphasize the utmost importance of preprocessing in augmenting the precision of sentiment classification. This research substantiates the system’s aptitude in extracting valuable insights for informed decision-making through the scrutiny of microblog sentiments.
  • Thumbnail Image
    Item type: Item ,
    DiagnoBot: A Medical Chatbot
    (2024-01-03) Nixon, Cheyenne; O’Barr, Benjamin; Gu, Keugmo
    Many people live without access to healthcare or delay care due to inconvenience, work, cost, living in rural areas, or social/medical fears (Gertz, Pollack, Schultheiss, & Brownstein, 2022), (Golembiewski, et al., 2022). Medical chatbots have emanated as a potential solution to healthcare access and to promote self-care. Our goal is to provide medical information through conversation to those who may otherwise delay seeking care. A Rasa chatbot is created using our Disease Prediction System, which utilizes machine learning algorithms i.e., Decision Trees, Gradient Boosting, Support Vector Machine (SVM), and Naïve Bayes to guide users to a sensible diagnosis, so they may opt to self-care at home or seek medical attention. In this paper, a sample of 4920 patient records with 41 disorders is analyzed. A Recursive Feature Elimination algorithm is used to enhance 95 out of the 132 symptom features. Our system achieved 97-100 percent accuracy.
  • Thumbnail Image
    Item type: Item ,
    Extending Feature Models with Types
    (2024-01-03) Callewaert, Benjamin; Vandevelde, Simon; Comenda, Nuno; Coppens, Bart; Decleyre, Nicholas; Vennekens, Joost
    Feature models are diagrams representing the variability of a product. While they are beneficial in reducing time, costs, and risks, their application in several problem areas remains unexplored. For example, in the financial domain, feature models could be leveraged to represent the variability of complex financial products and aid users in managing them. However, existing feature modeling approaches have limitations w.r.t. representing large products with complex features. To address some of these limitations, this work proposes a new approach called typed feature modeling, which extends the expressiveness of feature modeling by associating features with types that cover specified or infinite domains. We demonstrate the effectiveness of this approach by applying it to represent complex financial products that follow a commonly used industry standard. Additionally, we present an interactive tool for typed feature modeling containing an implementation of the financial use case and demonstrate how it can assist users in managing their financial products.
  • Thumbnail Image
    Item type: Item ,