Identify Topics from Text

No-Code Text Analysis

Texts may appear diverse at first glance, but many share underlying topics and thematic patterns.

In this tutorial, we introduce a method for identifying and extracting key topics from large volumes of text, allowing you to understand what themes structure the data as a whole.

This series introduces how to collect and analyze unstructured text data in NetMiner 5. ☞ Start NetMiner 5 Trial

What will you learn in this tutorial?

• How to identify key topics in text data

Who is this for?

• Researchers who want to uncover latent themes and discourses in text
• Researchers who want to go beyond frequency analysis to interpret semantic structure

Which NetMiner 5 feature is used?

• Topic Model Evaluation (Text > Topic Evaluation)
• Topic Modeling (Text > LDA)
• AI Assistant

In this tutorial, we use a sample dataset of AI academic publications. (File Location: \SampleData\Text Analysis\ENG\Bibliography) ☞ See Description of NetMiner sample data

1. Exploring the Right Topic Model for Your Data

Topic modeling is a method for summarizing large collections of text into a small number of topics.

LDA (Latent Dirichlet Allocation) is one of the most widely used approaches. It infers latent topic structures based on patterns of words that frequently appear together.

However, LDA requires the researcher to specify the number of topics in advance. If the number is too small, distinct themes may be merged into a single topic. If it is too large, topics can become overly fragmented. Because the results depend heavily on this choice, finding a topic count that fits the data is a critical step.

In NetMiner, you only need to define a range of topic numbers. The system evaluates which model best fits your data.

Go to Text > Topic Evaluation.

• If your input is raw text, select Raw Text.
• If your input is a bag-of-words representation, select TDM (Term-Document Matrix). (*If you select TDM, you can skip steps 1.1–1.3.)

On the right panel, choose the input data. In Data Item, select the dataset you imported. In Attribute, select the column that contains the unstructured text to analyze.

1.1 Morphological Analysis

Next, set the language for morphological analysis, choose the parts of speech to extract, and define the processing unit. For English, check Case Sensitive if you want to distinguish uppercase and lowercase words.

1.2 (Optional) User Dictionary

If you want to apply a user dictionary during the analysis, click the icon next to User Dictionary.

☞ Learn more about User Dictionaries

1.3 (Optional) Word Filtering

You can exclude unnecessary words using simple filters:

• Word Length: removes words that are too short
• Frequency: removes words that appear too rarely across documents
• TF-IDF: removes words that appear in almost all documents and carry low analytical value

1.4 Topic Model Generation Settings

NetMiner’s topic evaluation function generates multiple LDA topic models under different conditions and evaluates them using topic coherence and perplexity, helping you identify the model that best fits your data.

☞ Topic Coherence and Perplexity

In this step, set the conditions for topic model generation. Specify the minimum, maximum, and interval for number of topics (K), alpha, and eta. NetMiner will automatically generate and evaluate LDA models for all parameter combinations.

• Measure (Metric): choose the coherence metric. u_mass and c_v are available.
• # Top Words: number of top words per topic used to calculate model performance.
• # of Topics (K): range of topic numbers to evaluate.
• alpha / eta: LDA hyperparameters. alpha controls how evenly topics are distributed within documents. eta controls how evenly words are distributed within topics.
• # of Iterations: Maximum number of iterations through the corpus when inferring the topic distribution of a corpus.

2. Model Performance Results

Once the analysis is complete, NetMiner shows which model settings perform best based on coherence and perplexity.

In this example:

• Based on coherence, the best model has 7 topics with alpha/eta = 0.01.
• Based on perplexity, the best model has 5 topics, alpha = 0.02, and eta = 0.01.

Click the [C] Line Graph tab to visually inspect how model performance changes as the number of topics varies.

• For coherence, values closer to 0 (u_mass) or 1 (c_v) indicate better models.
• For perplexity, lower values indicate a better fit to the data.

3. Topic Modeling

Once you have identified the topic model settings that best fit your data, you can move on to extracting topics.

Go to Text > LDA. If your input data is raw text, select Raw Text. If you are using a word–document matrix (Bag-of-Words), select TDM.

In this tutorial, we use Raw Text.

The steps for selecting the input data, setting up morphological analysis, applying a user dictionary, and configuring word filtering are the same as in Step 1.

Finally, enter the topic model parameters identified in Step 2.

Click Run Process to start the analysis.

4. Topic Modeling Results

First, you will see a table listing the topics, numbered according to the selected number of topics. Each topic is shown with its top keywords, which helps you interpret the theme of the topic.

Below that, another table shows how many documents are assigned to each topic and their proportions in the corpus.

Click [C] LDAvis to explore the results visually.

On the left is the inter-topic distance map, where each circle represents a topic.

• Circle size: the proportion of the topic in the entire corpus
• Distance between circles: similarity between topics, based on keyword distributions. If circles overlap, the model may not be clearly separating those topics.

On the right, you will see the top keywords.

• Before selecting a topic, the panel shows globally frequent keywords.
• After selecting a topic, it shows keywords most strongly associated with that topic.

☞ Learn more about **how to interpret LDAvis**

5. Naming Topics with the AI Assistant

Click the AI Assistant button at the bottom of the report to get expert-level interpretations of your results, tailored to the context of your analysis.

※ Before using the AI Assistant, set your language model under Tools > Preferences. If you use a commercial LLM such as ChatGPT or Gemini (instead of a local Ollama model), you must register an API key from the provider.

☞ Configure language models in Preferences ☞ Get an API key for commercial LLMs

When the dialog opens, you may add a short description of your data (for example, “AI research articles”) to improve the interpretation. You can also proceed without input by clicking Generate Response.

AI Assistant Dialog

AI uses each topic’s representative keywords and proportions to suggest clear topic labels and provide an overall interpretation of the results.

By first comparing multiple topic models with Topic Model Evaluation and selecting the best-fitting one, you can run topic modeling with greater confidence and identify the latent thematic structure of your text more reliably.

Spend less time analyzing data. Spend more time on insight.

**☞ Try for Free ☞ NetMiner Help Center**

NetMiner | CYRAM Network Science · Social Network Analysis · Graph Analytics https://www.netminer.com