Instagram Text Analysis - FashionRec

Overview

This repository contains scripts and programs used to analyze textual data from the Instagram API in the context of automatic intelligent fashion identification, classification and recommendation.

Modules

• ./clean_data contains scripts for converting JSON Instagram data to csv,tsv,reduced json files that are suited for processing
• ./data_exploration contains some descriptive analytics about the data
• ./fasttext_on_spark contains a scalable implementation of FastText to run on Spark clusters
• ./information_extraction contains scripts for unsupervised information extraction using semantic/syntactic clustering of text to match it to an ontology/domain data as well as using several external APIs as sources of distant supervision. Also contains scripts for evaluation.
• ./wordvecs contains scripts training and evaluating word embeddings, as well as normalizing text corpora to be used for training word embeddings.
• ./cnn_classification contains scripts for training a weakly supervised CNN text classifier and model serving.

Usage

See individual README.md for each module

References

If using any of the code in this repsitory in your work or research, please cite: 1

Deep Text Mining of Instagram Data Without Strong Supervision

[1] K.Hammar, S.Jaradat, N.Dokoohaki, M.Matskin TODO

@INPROCEEDINGS{8609589,
author={K. Hammar and S. Jaradat and N. Dokoohaki and M. Matskin},
booktitle={2018 IEEE/WIC/ACM International Conference on Web Intelligence (WI)},
title={Deep Text Mining of Instagram Data without Strong Supervision},
year={2018},
volume={},
number={},
pages={158-165},
keywords={Twitter;Ontologies;Task analysis;Clothing;Text mining;Information Extraction;Instagram;Weak Supervision;Word Embeddings},
doi={10.1109/WI.2018.00-94},
ISSN={},
month={Dec},}

License

BSD 2-clause, see LICENSE

Authors/Contributors

Kim Hammar, kimham@kth.se

Shatha Jaradat, shatha@kth.se

Nima Dokoohaki, nimad@kth.se