A Reinforcement Learning Approach Combined With Scope Loss Function for Crime Prediction on Twitter (X)

Online social networks, especially Twitter (X), have become focal points for illicit activities, providing unique criminal investigation opportunities. This paper introduces an innovative methodology that uses social media sentiment analysis to predict criminal activities. One major challenge in sen...

Full description

Saved in:
Bibliographic Details
Main Authors: Liu Yang, Jiang Guofan, Zhang Yixin, Wei Qianze, Zhang Jian, Roohallah Alizadehsani, Pawel Plawiak
Format: Article
Language:English
Published: IEEE 2024-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10704669/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850282541568753664
author Liu Yang
Jiang Guofan
Zhang Yixin
Wei Qianze
Zhang Jian
Roohallah Alizadehsani
Pawel Plawiak
author_facet Liu Yang
Jiang Guofan
Zhang Yixin
Wei Qianze
Zhang Jian
Roohallah Alizadehsani
Pawel Plawiak
author_sort Liu Yang
collection DOAJ
description Online social networks, especially Twitter (X), have become focal points for illicit activities, providing unique criminal investigation opportunities. This paper introduces an innovative methodology that uses social media sentiment analysis to predict criminal activities. One major challenge in sentiment analysis is the uneven distribution of sentiment classes, where traditional models often fail to accurately classify instances of the minority class due to the overwhelming presence of majority class data. To tackle this issue, we propose a model that combines a reinforcement learning (RL) algorithm with a scope loss function. The RL approach uses a reward mechanism that assigns a more significant value to correctly predicting minority class instances over majority class ones. The scope loss function ensures an optimal balance between utilizing known data and exploring new data, thus maintaining a delicate equilibrium between accuracy and generalizability. Our model employs a series of convolutional neural networks (CNNs) to extract significant features from textual content, which are then utilized for sentiment classification. We also incorporate an advanced artificial bee colony (ABC) optimization technique to refine the model's hyperparameters. The effectiveness of our approach was empirically tested using two distinct datasets: one consisting of crime incident reports from the Chicago Police Department covering the period from September 2019 to July 2024 and another comprising tweets containing crime-related terms related to Chicago. The predictive capabilities of our proposed model were benchmarked against existing models, demonstrating superior performance with accuracies of 96.411% and 94.088%, respectively. This breakthrough highlights the potential of integrating sentiment analysis with reinforcement learning to significantly enhance the predictive accuracy of crime-related activities in online social networks, offering valuable insights for law enforcement and criminal investigation applications.
format Article
id doaj-art-00a76757a2ff456bbe402eff5bddc785
institution OA Journals
issn 2169-3536
language English
publishDate 2024-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-00a76757a2ff456bbe402eff5bddc7852025-08-20T01:47:58ZengIEEEIEEE Access2169-35362024-01-011214950214952710.1109/ACCESS.2024.347329610704669A Reinforcement Learning Approach Combined With Scope Loss Function for Crime Prediction on Twitter (X)Liu Yang0https://orcid.org/0009-0005-0853-5311Jiang Guofan1Zhang Yixin2Wei Qianze3Zhang Jian4https://orcid.org/0009-0002-2430-3818Roohallah Alizadehsani5https://orcid.org/0000-0003-0898-5054Pawel Plawiak6https://orcid.org/0000-0002-4317-2801Law School, Guangxi Police College, Nanning, ChinaDepartment of National Security, People's Public Security University of China, Beijing, ChinaBirmingham Law School, University of Birmingham, Birmingham, U.K.Department of Journalism and Communication, Hong Kong Chu Hai College, Tuen Mun, Hong KongSchool of Law, University of Malaya, Kuala Lumpur, MalaysiaInstitute for Intelligent Systems Research and Innovation (IISRI), Deakin University, Waurn Ponds, VIC, AustraliaDepartment of Computer Science, Faculty of Computer Science and Telecommunications, Cracow University of Technology, Krakow, PolandOnline social networks, especially Twitter (X), have become focal points for illicit activities, providing unique criminal investigation opportunities. This paper introduces an innovative methodology that uses social media sentiment analysis to predict criminal activities. One major challenge in sentiment analysis is the uneven distribution of sentiment classes, where traditional models often fail to accurately classify instances of the minority class due to the overwhelming presence of majority class data. To tackle this issue, we propose a model that combines a reinforcement learning (RL) algorithm with a scope loss function. The RL approach uses a reward mechanism that assigns a more significant value to correctly predicting minority class instances over majority class ones. The scope loss function ensures an optimal balance between utilizing known data and exploring new data, thus maintaining a delicate equilibrium between accuracy and generalizability. Our model employs a series of convolutional neural networks (CNNs) to extract significant features from textual content, which are then utilized for sentiment classification. We also incorporate an advanced artificial bee colony (ABC) optimization technique to refine the model's hyperparameters. The effectiveness of our approach was empirically tested using two distinct datasets: one consisting of crime incident reports from the Chicago Police Department covering the period from September 2019 to July 2024 and another comprising tweets containing crime-related terms related to Chicago. The predictive capabilities of our proposed model were benchmarked against existing models, demonstrating superior performance with accuracies of 96.411% and 94.088%, respectively. This breakthrough highlights the potential of integrating sentiment analysis with reinforcement learning to significantly enhance the predictive accuracy of crime-related activities in online social networks, offering valuable insights for law enforcement and criminal investigation applications.https://ieeexplore.ieee.org/document/10704669/Crime predictionsentiment analysisreinforcement learningclass imbalancehyperparameter tuningartificial bee colony
spellingShingle Liu Yang
Jiang Guofan
Zhang Yixin
Wei Qianze
Zhang Jian
Roohallah Alizadehsani
Pawel Plawiak
A Reinforcement Learning Approach Combined With Scope Loss Function for Crime Prediction on Twitter (X)
IEEE Access
Crime prediction
sentiment analysis
reinforcement learning
class imbalance
hyperparameter tuning
artificial bee colony
title A Reinforcement Learning Approach Combined With Scope Loss Function for Crime Prediction on Twitter (X)
title_full A Reinforcement Learning Approach Combined With Scope Loss Function for Crime Prediction on Twitter (X)
title_fullStr A Reinforcement Learning Approach Combined With Scope Loss Function for Crime Prediction on Twitter (X)
title_full_unstemmed A Reinforcement Learning Approach Combined With Scope Loss Function for Crime Prediction on Twitter (X)
title_short A Reinforcement Learning Approach Combined With Scope Loss Function for Crime Prediction on Twitter (X)
title_sort reinforcement learning approach combined with scope loss function for crime prediction on twitter x
topic Crime prediction
sentiment analysis
reinforcement learning
class imbalance
hyperparameter tuning
artificial bee colony
url https://ieeexplore.ieee.org/document/10704669/
work_keys_str_mv AT liuyang areinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT jiangguofan areinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT zhangyixin areinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT weiqianze areinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT zhangjian areinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT roohallahalizadehsani areinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT pawelplawiak areinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT liuyang reinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT jiangguofan reinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT zhangyixin reinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT weiqianze reinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT zhangjian reinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT roohallahalizadehsani reinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx
AT pawelplawiak reinforcementlearningapproachcombinedwithscopelossfunctionforcrimepredictionontwitterx