Using Text-Based Causal Inference to Disentangle Factors Influencing Online Review Ratings
Online reviews provide valuable insights into the perceived quality of facets of a product or service. While aspect-based sentiment analysis has focused on extracting these facets from reviews, there is less work understanding the impact of each aspect on overall perception. This is particularly cha...
I tiakina i:
| Ngā kaituhi matua: | , , |
|---|---|
| Hōputu: | Tuhinga |
| I whakaputaina: |
2026
|
| Ngā marau: | |
| Urunga tuihono: | https://arxiv.org/abs/2606.04286 http://arxiv.org/abs/2606.04286 |
| Ngā Tūtohu: |
Kāore He Tūtohu, Me noho koe te mea tuatahi ki te tūtohu i tēnei pūkete!
|
| _version_ | 1879179178775937024 |
|---|---|
| author | Li, Linsen Culotta, Aron Mattei, Nicholas |
| author_browse | Culotta, Aron Li, Linsen Mattei, Nicholas |
| author_facet | Li, Linsen Culotta, Aron Mattei, Nicholas |
| author_sort | Li, Linsen |
| building | arXiv |
| collection | Acceso abierto |
| description | Online reviews provide valuable insights into the perceived quality of facets of a product or service. While aspect-based sentiment analysis has focused on extracting these facets from reviews, there is less work understanding the impact of each aspect on overall perception. This is particularly challenging given correlations among aspects, making it difficult to isolate the effects of each. This paper introduces a methodology based on recent advances in text-based causal analysis, specifically CausalBERT, to disentangle the effect of each factor on overall review ratings. We enhance CausalBERT with three key improvements: temperature scaling for better calibrated treatment assignment estimates; hyperparameter optimization to reduce confound overadjustment; and interpretability methods to characterize discovered confounds. In this work, we treat the textual mentions in reviews as proxies for real-world attributes. We validate our approach on real and semi-synthetic data from over 600K reviews of U.S. K-12 schools. We find that the proposed enhancements result in more reliable estimates, and that perception of school administration and performance on benchmarks are significant drivers of overall school ratings. |
| doi_str_mv | 10.18653/v1/2025.naacl-long.562 |
| first_indexed | 2026-10-16T01:28:34Z |
| format | Article |
| fullrecord | <oai_dc:dc xmlns:oai_dc="http://www.openarchives.org/OAI/2.0/oai_dc/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/oai_dc/ https://www.openarchives.org/OAI/2.0/oai_dc.xsd"><identifier>arxiv-2606.04286</identifier><datestamp>2026-06-04</datestamp>
<dc:title>Using Text-Based Causal Inference to Disentangle Factors Influencing Online Review Ratings</dc:title>
<dc:creator>Li, Linsen</dc:creator>
<dc:creator>Culotta, Aron</dc:creator>
<dc:creator>Mattei, Nicholas</dc:creator>
<dc:subject>Computation and Language</dc:subject>
<dc:description>Online reviews provide valuable insights into the perceived quality of facets of a product or service. While aspect-based sentiment analysis has focused on extracting these facets from reviews, there is less work understanding the impact of each aspect on overall perception. This is particularly challenging given correlations among aspects, making it difficult to isolate the effects of each. This paper introduces a methodology based on recent advances in text-based causal analysis, specifically CausalBERT, to disentangle the effect of each factor on overall review ratings. We enhance CausalBERT with three key improvements: temperature scaling for better calibrated treatment assignment estimates; hyperparameter optimization to reduce confound overadjustment; and interpretability methods to characterize discovered confounds. In this work, we treat the textual mentions in reviews as proxies for real-world attributes. We validate our approach on real and semi-synthetic data from over 600K reviews of U.S. K-12 schools. We find that the proposed enhancements result in more reliable estimates, and that perception of school administration and performance on benchmarks are significant drivers of overall school ratings. </dc:description>
<dc:description>HLT/NAACL 2025</dc:description>
<dc:date>2026-06-02</dc:date>
<dc:type>text</dc:type>
<dc:identifier>https://arxiv.org/abs/2606.04286</dc:identifier>
<dc:identifier>http://arxiv.org/abs/2606.04286</dc:identifier>
<dc:identifier>In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies</dc:identifier>
<dc:identifier>doi:10.18653/v1/2025.naacl-long.562</dc:identifier>
</oai_dc:dc> |
| id | arxiv-2606.04286 |
| institution | arXiv |
| last_indexed | 2026-10-16T01:28:34Z |
| publishDate | 2026 |
| publishDateRange | 2026 |
| publishDateSort | 2026 |
| record_format | oai_dc |
| spelling | arxiv-2606.042862026-06-04 Using Text-Based Causal Inference to Disentangle Factors Influencing Online Review Ratings Li, Linsen Culotta, Aron Mattei, Nicholas Computation and Language Online reviews provide valuable insights into the perceived quality of facets of a product or service. While aspect-based sentiment analysis has focused on extracting these facets from reviews, there is less work understanding the impact of each aspect on overall perception. This is particularly challenging given correlations among aspects, making it difficult to isolate the effects of each. This paper introduces a methodology based on recent advances in text-based causal analysis, specifically CausalBERT, to disentangle the effect of each factor on overall review ratings. We enhance CausalBERT with three key improvements: temperature scaling for better calibrated treatment assignment estimates; hyperparameter optimization to reduce confound overadjustment; and interpretability methods to characterize discovered confounds. In this work, we treat the textual mentions in reviews as proxies for real-world attributes. We validate our approach on real and semi-synthetic data from over 600K reviews of U.S. K-12 schools. We find that the proposed enhancements result in more reliable estimates, and that perception of school administration and performance on benchmarks are significant drivers of overall school ratings. HLT/NAACL 2025 2026-06-02 text https://arxiv.org/abs/2606.04286 http://arxiv.org/abs/2606.04286 In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies doi:10.18653/v1/2025.naacl-long.562 |
| spellingShingle | Computation and Language Li, Linsen Culotta, Aron Mattei, Nicholas Using Text-Based Causal Inference to Disentangle Factors Influencing Online Review Ratings |
| title | Using Text-Based Causal Inference to Disentangle Factors Influencing Online Review Ratings |
| title_full | Using Text-Based Causal Inference to Disentangle Factors Influencing Online Review Ratings |
| title_fullStr | Using Text-Based Causal Inference to Disentangle Factors Influencing Online Review Ratings |
| title_full_unstemmed | Using Text-Based Causal Inference to Disentangle Factors Influencing Online Review Ratings |
| title_short | Using Text-Based Causal Inference to Disentangle Factors Influencing Online Review Ratings |
| title_sort | using text based causal inference to disentangle factors influencing online review ratings |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2606.04286 http://arxiv.org/abs/2606.04286 |
| work_keys_str_mv | AT lilinsen usingtextbasedcausalinferencetodisentanglefactorsinfluencingonlinereviewratings AT culottaaron usingtextbasedcausalinferencetodisentanglefactorsinfluencingonlinereviewratings AT matteinicholas usingtextbasedcausalinferencetodisentanglefactorsinfluencingonlinereviewratings |