Scribd
Upload
7 views2 pages

Text Clustering and Topic Modeling Insights

The document discusses two methods for text clustering: Co-clustering and Topic Modeling (PLSA). Co-clustering simultaneously clusters documents and words to provide clearer thematic insights, while PLSA uses a probabilistic approach to uncover latent topics, allowing documents to be associated with multiple topics. Co-clustering results in non-overlapping clusters, whereas PLSA offers soft clustering and dimensionality reduction.

Uploaded by

rehanaparveen479
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
7 views2 pages

Text Clustering and Topic Modeling Insights

The document discusses two methods for text clustering: Co-clustering and Topic Modeling (PLSA). Co-clustering simultaneously clusters documents and words to provide clearer thematic insights, while PLSA uses a probabilistic approach to uncover latent topics, allowing documents to be associated with multiple topics. Co-clustering results in non-overlapping clusters, whereas PLSA offers soft clustering and dimensionality reduction.

Uploaded by

rehanaparveen479
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Text Clustering and Topic Modeling - Summary Notes

1. Simultaneous Document and Word Clustering (Co-clustering)

- In text mining, data is represented as a high-dimensional sparse matrix (document-term matrix).

- Clustering only documents is often not enough to understand the topics of those clusters.

- Co-clustering solves this by clustering documents and words simultaneously.

Method:

- Documents and words are represented as nodes in a bipartite graph.

- An edge between a document and a word represents term frequency.

- The goal is to partition the graph such that each cluster has a group of related documents and

topical words.

- This rearranges the matrix into blocks, making clusters visually and semantically clearer.

- Each document and word belongs to exactly one cluster (non-overlapping).

Advantages:

- Provides both document clusters and meaningful word clusters.

- Helps interpret the themes of each cluster more effectively.

2. Topic Modeling (PLSA - Probabilistic Latent Semantic Analysis)

- PLSA is a probabilistic approach to uncover latent topics in a collection of documents.

- It can be seen as a probabilistic version of LSA (Latent Semantic Analysis).

How it works:

1. Assume there are k latent topics (G1, G2, ..., Gk).


2. For each document-word pair, choose a topic Gm with probability P(Gm).

3. Generate document and word independently with P(doc | Gm) and P(word | Gm).

4. Increase the frequency count for the corresponding matrix entry.

Characteristics:

- A document can be generated from multiple topics (soft clustering).

- It reduces dimensionality and extracts interpretable topics.

- Closely related to nonnegative matrix factorization (NMF).

Comparison:

- Co-clustering yields hard clusters for both documents and words.

- PLSA produces soft clusters and focuses more on reducing dimensionality.

You might also like