A tree based keyphrase extraction technique for academic literature
Automatic keyphrase extraction techniques aim to extract quality keyphrases to summarize a document at a higher level. Among the existing techniques some of them are domain-specific and require application domain knowledge, some of them are based on higher-order statistical methods and are computati...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://umpir.ump.edu.my/id/eprint/32740/1/A%20tree%20based%20keyphrase%20extraction%20technique%20for%20academic%20literature.wm.pdf http://umpir.ump.edu.my/id/eprint/32740/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Universiti Malaysia Pahang |
| Language: | English |
| Summary: | Automatic keyphrase extraction techniques aim to extract quality keyphrases to summarize a document at a higher level. Among the existing techniques some of them are domain-specific and require application domain knowledge, some of them are based on higher-order statistical methods and are computationally expensive, and some of them require large train data which are rare for many applications. Overcoming these issues, this thesis proposes a new unsupervised automatic keyphrase extraction technique, named TeKET or Tree-based Keyphrase Extraction Technique, which is domain-independent, employs limited statistical knowledge, and requires no train data. The proposed technique also introduces a new variant of the binary tree, called KeyPhrase Extraction (KePhEx) tree to extract final keyphrases from candidate keyphrases. Depending on the candidate keyphrases the KePhEx tree structure is either expanded or shrunk or maintained. In addition, a measure, called Cohesiveness Index or CI, is derived that denotes the degree of cohesiveness of a given node with respect to the root which is used in extracting final keyphrases from a resultant tree in a flexible manner and is utilized in ranking keyphrases alongside Term Frequency. The effectiveness of the proposed technique is evaluated using an experimental evaluation on a benchmark corpus, called SemEval-2010 with total 244 train and test articles, and compared with other relevant unsupervised techniques by taking the representatives from both statistical (such as Term Frequency-Inverse Document Frequency and YAKE) and graph-based techniques (PositionRank, CollabRank (SingleRank), TopicRank, and MultipartiteRank) into account. Three evaluation metrics, namely precision, recall and F1 score are taken into consideration during the experiments. The obtained results demonstrate the improved performance of the proposed technique over other similar techniques in terms of precision, recall, and F1 scores. |
|---|