Source-linked AI summary
Unsupervised Keyphrase Extraction with Multipartite Graphs
Florian Boudin
TL;DR
Keyphrase extraction needs unsupervised methods that better handle topical coverage, diversity, and representative candidate selection. The paper uses a multipartite graph to jointly model topics and candidates, combines their mutually reinforcing relations with selection preferences, and reports improved performance over graph-based baselines across three datasets.
Problem
Existing graph-based methods have difficulty ensuring topical coverage and diversity, while topic-aware and topic-level approaches face computational, data, independence, or representative-selection limitations.
Method
An unsupervised multipartite graph represents topic-related candidate sets and ranks candidates in one operation, with edge-weight adjustments incorporating selection preferences.
Results
The model achieves the best overall results and significantly outperforms baselines on most metrics across three public datasets.
Takeaways & Limitations
Over 92% of the top-10 extracted keyphrases are assigned to different topics, indicating that the model promotes diversity without hard constraints.
Takeaways & Limitations
Performance drops sharply for F1@5 on Semeval-2010, and the authors hypothesize difficulty selecting representative candidates when TextRank is the sole feature.
Abstract
from arXiv · showhide
We propose an unsupervised keyphrase extraction model that encodes topical information within a multipartite graph structure. Our model represents keyphrase candidates and topics in a single graph and exploits their mutually reinforcing relationship to improve candidate ranking. We further introduce a novel mechanism to incorporate keyphrase selection preferences into the model. Experiments conducted on three widely used datasets show significant improvements over state-of-the-art graph-based models.
1 Introduction
Graph-based keyphrase extraction is attractive because it is unsupervised, but conventional graph designs struggle to represent topical coverage and diversity. The proposed multipartite graph jointly represents topics and candidates, enabling a single ranking operation that exploits their mutually reinforcing relationship and incorporates selection preferences.
- Motivation: Graph-based approaches build document graphs and rank their nodes, offering strong performance while remaining completely unsupervised.Typical graphs use lexical units as nodes and semantic relations as edges.
- Motivation: Conventional graph representations do not naturally ensure topical coverage and diversity, a central challenge in keyphrase extraction.Prior work largely focused on semantic relatedness or node-ranking functions, with limited attention to graph types.
- Motivation: Topic-biased ranking models can be computationally prohibitive for long documents, require large datasets for word-topic distributions, and assume independent topics.These limitations make topic diversity difficult to ensure without additional supervised approaches.
- Prior approaches: TopicRank improves ranking and topic coverage but treats candidates within one topic as equally important, requiring post-ranking heuristics for representative selection.Its topic-level representation can also propagate errors from topic formation.
- Proposed approach: The proposed multipartite graph encodes candidates in topic-based independent sets and ranks them in one operation, using mutually reinforcing topic-candidate relations.The representation is illustrated against TopicRank for a Hulth-2003 document; the supplied figure caption identifies nodes and co-occurrence edges.
- Proposed approach: A selection-preference mechanism adjusts edge weights to promote candidates with properties such as thesaurus matches or early document occurrence.The introduction specifically demonstrates a bias toward candidates occurring first in the document.
2 Proposed Model
The model represents topic-related keyphrase candidates in a directed multipartite graph, adjusts edge weights to encode selection preferences, and ranks candidates with weighted TextRank.
- Candidate and topic construction: Candidates are extracted from adjacent adjective–noun sequences and grouped into topics by hierarchical agglomerative clustering over shared word stems.The setup follows Bougouin et al. (2013) for direct comparability.
- Graph construction: Candidate edge weights sum inverse distances between their document occurrences, matching window-based co-occurrence performance without parameter tuning.P(ci) contains the word-offset positions of candidate ci.
- Multipartite graph representation: The complete k-partite graph connects candidates only across different topics, while each topic forms an independent set of unconnected nodes.Here, k denotes the number of topics.
- Multipartite graph representation: Removing intra-topic edges distributes recommendation across topics and implicitly promotes topical diversity through mutually reinforcing topic–candidate relations.The representation does not assume a particular topic-decomposition method.
- Graph weight adjustment mechanism: The model adjusts incoming edges of each topic’s first-occurring candidate, promoting candidates according to other candidates in the same topic.The selection heuristic can instead prioritize properties such as thesaurus membership.
- Ranking and extraction: Weighted TextRank orders candidates by relevance, after which the top N candidates are selected as keyphrases.The damping factor λ is set to 0.85, although other ranking algorithms can be applied.
3 Experiments
Experiments evaluate the unsupervised model on three datasets using top-N F1 and MAP, against three graph-based baselines. The model achieves the best overall results, promotes topical diversity, and exposes limitations in representative-candidate selection.
- 3.1 Datasets and evaluation measures: Performance is measured by F1 at the top 5 and 10 keyphrases and by Mean Average Precision (MAP), with stemming applied before scoring.Significance markers in Table 1 indicate the 0.05 level using Student’s t-test.
- 3.2 Baselines and parameter settings: The model is compared with TopicRank, Single Topical PageRank, and PositionRank, covering topic-based and feature-enhanced graph-ranking approaches.LDA topic distributions are estimated on each dataset’s training set, and the proposed model tunes graph-adjustment strength using α.
- 3.3 Results: Overall, the proposed model achieves the best results and significantly outperforms the baselines on most metrics, with smaller relative gains on the shorter Hulth-2003 and Marujo-2012 documents.TopicRank has the highest baseline precision, while PositionRank is the strongest baseline by MAP.
- 3.3 Results: Without weight adjustment, results remain higher or on par with topic-informed baselines, supporting the model’s use of reinforcing topic–candidate relations.The performance drop is especially severe for F1@5 on SemEval-2010, where the model moves from best to worst performance.
- 3.3 Results: Over 92% of the top-10 keyphrases are assigned to different topics, indicating topical diversity without hard constraints.Some apparent redundancy reflects clustering errors or hypernym–hyponym pairs that both occur in the gold references.
4 Conclusion
The paper concludes that its unsupervised multipartite-graph model is effective across three public datasets. It also releases code and data and identifies structure-specific ranking algorithms as future work.
- The unsupervised multipartite-graph model is demonstrated on three public datasets.
- The authors make their code and data publicly available through the project repository.
- Future work will apply ranking algorithms that leverage the specific structure of the multipartite graph.