Icon

Topic Models from reviews

Topic Models from Reviews

This workflow addresses the problem of extracting and modeling topics from reviews.

Block 1 performs the data preparation on review texts. Block 2 optimizes the parameters for the LDA algorithm. Block 3 applies the LDA algorithm with optimized parameters and displays the LDA topic probabilities along with the average number of stars by topic. Block 4 estimates the importance of topics via linear regression (KNIME) and polynomial regression (R).

If you use this workflow, please cite:
F. Villaroel Ordenes & R. Silipo, “Machine learning for marketing on the KNIME Hub: The development of a live repository for marketing applications”, Journal of Business Research 137(1):393-410, DOI: 10.1016/j.jbusres.2021.08.036.

URL: 10.1016/j.jbusres.2021.08.036 http://10.1016/j.jbusres.2021.08.036

Block 2 - Find optimal k for topic models: This block finds the optimal k for the LDA topic modeling algorithm. Other methods for topic extraction and modeling can be implemented in KNIME (see https://hub.knime.com/angusveitch/spaces/Public/latest/TopicKR~HRMp6v9Ip_ODMIob). Other topic modeling algorithms that can be used in R or Python are structural topic models (STM) and correlated topic models (CTM).
Block 4 - Analysis to inspect the impact of topics on customer star rating: Analysis can be improved by including topic sentiment, interaction terms, and different modeling alternatives (e.g., ordinal logit regression in R).
Block 1 - Data Preparation for Topic Models: This block performs preprocessing, extraction of n-grams, and exclusion of reviews with a small number of terms. It can be adjusted as desired
Block 3 - Obtain topic solution: Users can test more than one solution for topic extraction and choose the best one based on interpretability. The "topic analysis component" needs to be manually edited to rename topics if changes are made at any earlier stage of the process.
Analysis of customer experience feedback with topic models The study of customer experience management (CXM) with big data analytics (BDA) is one of the most relevant marketing analytics topics in the last years. The present workflow shows how managers can identify service aspects with a greater impact on customer overall evaluation (star rating). The workflow shows as well how to integrate R for statistical analysis within KNIME
Star_Rating
Number to String
- DV: Star Rating - IV's Topic Probabilities baseline: Topic 0 - Control: Hotel
Linear Regression Learner
Summary words per topic
GroupBy
TripAdvisor 2 hotels
Excel Reader
Step 3: Select a topic
Topic Extractor (Parallel LDA)
Doc Creation
Topic Comparison Per Hotel
Topic Analysis
Preprocessing
Visualize Perplexity
bi-grams (tri-grams can be added as well)
N-grams
Assessmodel fit (perplexity), in a wide range of topics (2 to 80). Identify elbow range in image
Step 1: Optimal k in [2,80]
avg # stars by topic per hotel
Topic Analysis
Narrow down the search on a smaller range from step 1. Identify elbow point (15)
Step 2: Optimal k in [2,20]
Filter Reviews with Less than 10 words

Nodes

Extensions

Links