Disentangling Political Bias in Social Media Data through Multi-task Embedding Learning (Under Review)

An implementation of DisMEL+TransE and Data used in the paper [link]. Run experiment by the following command

python dismel_transe.py

Tested on python==3.8.8 and tensorflow==2.6.0

jp-poli_rel_text.txt is the graph data represented as (head, relation, tail) triplets. tweet_text.txt contains tweet text and tweet id in the graph file.

Extracting Political Interest Model from Interaction Data Based on Novel Word-level Bias Assignment (TIST2025)

Code for BiDEL [link].

This package contains code and data for inferring word-level political bias from labeled texts. Please see readme.txt in the package for details.

Word-Level Political Sentiments Inferred from Social Media and Application in Recommendation Diversification (TWEB2024)

A Political Sentiment Dictionary for Japanese Language [html] [text].

This dictionary contains interest scores and political sentiment scores of 63,588 words, generated using the method presented in the paper. The TSV file contains three columns, for the word, interest score, and political sentiment score, respectively. A higher value indicates a leaning towards left, while a lower value indicates a leaning towards right.

Event Embedding Learning from Social Media Using Graph Topic Model Autoencoder (ASONAM 2024)

Code and data for repeating the Reseed experiment in the paper. [link]

Evolving Social Media Background Representation with Frequency Weights and Co-occurrence Graphs (TKDD2023)

Source code written in Python and TensorFlow [link].

The package contains an implementation of frequency-weighted GCN (FW-GCN). It also contains intermediate data for the first scenario, tweet popularity prediction. Two baseline methods are provided for comparison, MF and GCN.

Generalized Durative Event Detection on Social Media (JIIS 2022)
A General Method for Event Detection on Social Media (ADBIS 2020)

Codes written in R [link].

Predicting Time and Location of Future Crimes with Recommendation Methods (Knowledge-Based Systems 2020)

Data and source codes written in R [link].

Automatic Latent Street Type Discovery from Web Open Data (Information Systems 2020)

Data and source codes written in Java [link]. The codes include a Java implementation of WM-LDA and a program for testing coherence and association described in the paper.

Analysis of Street Crime Predictors in Web Open Data (JIIS 2019)

Data used in the paper can be downloaded here.

Image Tweet Popularity Prediction with Convolutional Neural Network (ECIR'19)

This python program uses tensorflow to construct the convolutional neural network described in the paper.