Graph Based Approach to Text Mining Solutions
summary
Method demonstrates discovering new senses for Slovenian collocations by building a knowledge graph from a CLARIN.SI collocation table (~1,700 rows yielding ~2,400 distinct nodes and ~1,700 edges, many small components), computing degree/betweenness/closeness and noting a chemical-element node with highest centrality, then improving with node embeddings and clustering. Implementation uses NetworkX for graphs, node2vec + t-SNE for 2D embeddings, and gensim LDA for topical signals; centrality-based sense prediction works but is sparse, embeddings plus clustering give clearer semantic assignments.
AI-written summary · about
chapters (11)
- 0:00Graph-based text mining intro
- 1:54Data setup and graph theory
- 14:09Graph creation in Python
- 24:12Centrality measures calculation
- 39:52Degree centrality for sense prediction
- 48:06Shortest paths between collocations
- 53:57Node2Vec embeddings
- 1:03:27Collocation clustering with K-means & DBSCAN
- 1:10:16Advanced graph embeddings
- 1:20:52LDA topic modeling
- 1:28:45LDA keyword extraction
related talks
33mSmarter Industrial Fault Diagnosis with Knowledge Graphs and Large Language Models
28mBuilding Knowledge Graph-Based Agents with Structured Text Generation
30mMaking Virtuous Cycles: Lessons From the Frontier of LLM-Knowledge Graph Interaction
31mBeyond Vectors: Evolving GenAI through Transformative Tools and Methods
46mKnowledge Graphs + LLMs: What's the latest in research?
30mFrom Unstructured Docs to Graph Intelligence: A Framework for Building Graph+Embedding Pipelines
22mBuilding Contextual Knowledge Graphs to Adapt, Evaluate, and Leverage LLMs
42mMaking OSS AI Real with Knowledge Graphs