|
Identifier |
000421544 |
Title |
On achieving high quality user reviews retrieval in the context of conversational faceted search |
Alternative Title |
Υψηλής ποιότητας ανάκτηση σχολίων χρηστών στα πλαίσια της διαλογικής εξερευνητικής αναζήτησης |
Author
|
Σγόντζος, Κωνσταντίνος Α.
|
Thesis advisor
|
Τζίτζικας, Γιάννης
|
Reviewer
|
Πλεξουσάκης, Δημήτρης
Φλουρής, Γεώργιος
|
Abstract |
Question Answering Systems can be considered as an extension of Search Engines
in the sense that they directly answer questions posed by users in Natural Language
instead of simply returning relevant sources based on keyword search. Recently, such
systems gained more attention and have great potential in various application domains
including e-commerce.
The general context of our work is to support Faceted Search (FS) in the form of spoken
dialogue. FS is an exploratory search paradigm that is commonly applied over
multidimensional or graph data. However, since structured data might not suffice for
answering a user’s query, unstructured information like user comments or reviews can
be a valuable source of information. In this work we focus on how to leverage the textual
information (user reviews and comments) for answering the user questions. This
information is generally restricted in the length of a sentence or a paragraph and is known
as Answer Passage (AP). Due to the speech interaction with the user, high quality user
review retrieval is necessary.
To study this problem, we created an evaluation collection and we adjusted to our
problem an existing TREC collection. We then proposed three basic approaches and
variations of them. These approaches are based on handmade (Wordnet) and statistical
(Word2vec) dictionaries. The comparative experimental results show that: (a) The
combination of handmade and statistical dictionaries succeed better results in terms of
precision in relevant AP (review) retrieval, (b) The expansion of the question with other
relevant terms combined with methods that devalue question words which describe the
general domain context, improve the precision, mainly in restricted domains such as hotel
booking. In comparison to existing related work, our proposed approaches achieve higher
precision.
|
Language |
English |
Subject |
Faceted search |
|
User reviews |
|
Word embeddings |
|
Wordnet |
|
Πολυεδρική αναζήτηση |
|
Στατιστικά λεξικά |
|
Σχόλια χρηστών |
|
Χειρόγραφα λεξικά |
Issue date |
2019-03-29 |
Collection
|
School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses
|
|
Type of Work--Post-graduate theses
|
Views |
591 |