ECIR2023_Multimodal Complaint-Detection

Knowing What and How: A Multi-modal Aspect-Based Framework for Complaint Detection

This repository contains the resource files for our work "Knowing What and How: A Multi-modal Aspect-Based Framework for Complaint Detection", accepted in ECIR 2023.

CESAMARD Dataset

The original CESAMARD dataset released in Sentiment and Emotion-aware Multi-modal Complaint Identification. The CESAMARD dataset is a multimodal complaint corpus (text and images) for research in automated complaint discovery. The dataset is compiled from ecommerce webiste Amazon. CESAMARD consists of 3962 reviews with the corresponding complaint, emotion, and sentiment labels.

CESAMARD-Aspect Dataset

The CESAMARD-Aspect dataset is an extended version of CESAMARD dataset. The extended dataset consists of 3962 multimodal complaint reviews (texts and images) in English. Each record in the dataset consists of review text, review image (user-uploaded) domain, review title, complaint labels (overall) and corresponding annotations for aspect category, aspect-level complaint/non-complaint labels.

CITE

@inproceedings{singh2023knowing, title={Knowing what and how: a multi-modal aspect-based framework for complaint detection}, author={Singh, Apoorva and Gangwar, Vivek and Sharma, Shubham and Saha, Sriparna}, booktitle={European Conference on Information Retrieval}, pages={125--140}, year={2023}, organization={Springer} }