dc.contributor.author | Cui, Jiarong | |
dc.date.accessioned | 2021-08-30T18:33:54Z | |
dc.date.available | 2021-08-30T18:33:54Z | |
dc.date.issued | 2021-08-30T18:33:54Z | |
dc.identifier.uri | http://hdl.handle.net/10222/80763 | |
dc.description | eXSTS comprises two parts, the back-end data analysis system and the front-end user interface (UI). The back-end system consists of a Siamese Neural Network (SNN) and BERT\textsubscript{base} pre-trained model. eXSTS receives two documents, and the back-end system splits two documents into sentences, defines pairs of sentences across the two documents, and calculates the STS score of each sentence pair by comparing the sentence embeddings of the two sentences. We normalize all the STS scores and calculate the document STS score of these two documents. The front-end UI visualizes the STS information of sentence pairs through proposed visualizations to explain the sentence pairs that significantly affect the document STS score of two documents and the global distribution of the STS scores of all the sentences pairs across two documents. | en_US |
dc.description.abstract | We propose eXplainable Semantic Textual Similarity (eXSTS) with two visualizations that allow the users to investigate the Semantic Textual Similarity (STS) relationship between two documents. eXSTS offers insights for the users who are not familiar with Natural Language Processing to the STS relationship between two documents. eXSTS was invented to deal with the job advertisement classification task. When a job advertisement is entered into eXSTS, eXSTS retrieves the five most relevant National Occupational Classification (NOC) unit group based on the document STS score of the job advertisement and each NOC unit group. The front-end user interface demonstrates the STS relationship between the job advertisement and one NOC unit group in the five most relevant NOC unit groups to explore the important STS information across the job advertisement and this NOC unit group and why eXSTS chose this NOC unit group to opt-in the five most relevant NOC unit groups. | en_US |
dc.language.iso | en | en_US |
dc.subject | Information Retrieval | en_US |
dc.subject | Semantic Textual Similarity | en_US |
dc.subject | Human-computer interaction | en_US |
dc.subject | Visualization | en_US |
dc.subject | Job advertisement classification | en_US |
dc.title | eXSTS: eXplainable Semantic Textual Similarity | en_US |
dc.type | Thesis | en_US |
dc.date.defence | 2021-08-20 | |
dc.contributor.department | Faculty of Computer Science | en_US |
dc.contributor.degree | Master of Computer Science | en_US |
dc.contributor.external-examiner | n/a | en_US |
dc.contributor.graduate-coordinator | Michael McAllister | en_US |
dc.contributor.thesis-reader | VLADO KESELJ | en_US |
dc.contributor.thesis-reader | FERNANDO PAULOVICH | en_US |
dc.contributor.thesis-supervisor | EVANGELOS E. MILIOS | en_US |
dc.contributor.ethics-approval | Received | en_US |
dc.contributor.manuscripts | No | en_US |
dc.contributor.copyright-release | No | en_US |