JLJoyce Liu

Joyce Liu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Joyce Liu has a total of 15 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Tracking and Profiling Repeated Users Over Time in Text-Based Counseling: Longitudinal Observational Study With Hierarchical Clustering

    by Yucan Xu, Christian Shaunlyn Chan, Evangeline Chan, Junyou Chen, Florence Cheung, Zhongzhi Xu, Joyce Liu, Paul Siu Fai Yip

    Background: Due to their accessibility and anonymity, web-based counseling services are expanding at an unprecedented rate. One of the most prominent challenges such services face is repeated users, who represent a small fraction of total users but consume significant resources by continually returning to the system and reiterating the same narrative and issues. A deeper understanding of repeated users and tailoring interventions may help improve service efficiency and effectiveness. Previous studies on repeated users were mainly on telephone counseling, and the classification of repeated users tended to be arbitrary and failed to capture the heterogeneity in this group of users. Objective: In this study, we aimed to develop a systematic method to profile repeated users and to understand what drives their use of the service. By doing so, we aimed to provide insight and practical implications that can inform the provision of service catering to different types of users and improve service effectiveness. Methods: We extracted session data from 29,400 users from a free 24/7 web-based counseling service from 2018 to 2021. To systematically investigate the heterogeneity of repeated users, hierarchical clustering was used to classify the users based on 3 indicators of service use behaviors, including the duration of their user journey, use frequency, and intensity. We then compared the psychological profile of the identified subgroups including their suicide risks and primary concerns to gain insights into the factors driving their patterns of service use. Results: Three clusters of repeated users with clear psychological profiles were detected: episodic, intermittent, and persistent-intensive users. Generally, compared with one-time users, repeated users showed higher suicide risks and more complicated backgrounds, including more severe presenting issues such as suicide or self-harm, bullying, and addictive behaviors. Higher frequency and intensity of service use were also associated with elevated suicide risk levels and a higher proportion of users citing mental disorders as their primary concerns. Conclusions: This study presents a systematic method of identifying and classifying repeated users in web-based counseling services. The proposed bottom-up clustering method identified 3 subgroups of repeated users with distinct service behaviors and psychological profiles. The findings can facilitate frontline personnel in delivering more efficient interventions and the proposed method can also be meaningful to a wider range of services in improving service provision, resource allocation, and service effectiveness.


  • Top Cited Papers
    Open Access

    Efficacy of ChatGPT in Cantonese Sentiment Analysis: Comparative Study

    by Ziru Fu, Yu Cheng Hsu, Christian S Chan, Chaak Ming Lau, Joyce Liu, Paul Siu Fai Yip

    Background: Sentiment analysis is a significant yet difficult task in natural language processing. The linguistic peculiarities of Cantonese, including its high similarity with Standard Chinese, its grammatical and lexical uniqueness, and its colloquialism and multilingualism, make it different from other languages and pose additional challenges to sentiment analysis. Recent advances in models such as ChatGPT offer potential viable solutions. Objective: This study investigated the efficacy of GPT-3.5 and GPT-4 in Cantonese sentiment analysis in the context of web-based counseling and compared their performance with other mainstream methods, including lexicon-based methods and machine learning approaches. Methods: We analyzed transcripts from a web-based, text-based counseling service in Hong Kong, including a total of 131 individual counseling sessions and 6169 messages between counselors and help-seekers. First, a codebook was developed for human annotation. A simple prompt (“Is the sentiment of this Cantonese text positive, neutral, or negative? Respond with the sentiment label only.”) was then given to GPT-3.5 and GPT-4 to label each message’s sentiment. GPT-3.5 and GPT-4’s performance was compared with a lexicon-based method and 3 state-of-the-art models, including linear regression, support vector machines, and long short-term memory neural networks. Results: Our findings revealed ChatGPT’s remarkable accuracy in sentiment classification, with GPT-3.5 and GPT-4, respectively, achieving 92.1% (5682/6169) and 95.3% (5880/6169) accuracy in identifying positive, neutral, and negative sentiment, thereby outperforming the traditional lexicon-based method, which had an accuracy of 37.2% (2295/6169), and the 3 machine learning models, which had accuracies ranging from 66% (4072/6169) to 70.9% (4374/6169). Conclusions: Among many text analysis techniques, ChatGPT demonstrates superior accuracy and emerges as a promising tool for Cantonese sentiment analysis. This study also highlights ChatGPT’s applicability in real-world scenarios, such as monitoring the quality of text-based counseling services and detecting message-level sentiments in vivo. The insights derived from this study pave the way for further exploration into the capabilities of ChatGPT in the context of underresourced languages and specialized domains like psychotherapy and natural language processing.


  • Open Access

    Efficacy of ChatGPT in Cantonese Sentiment Analysis: Comparative Study (Preprint)

    by Ziru Fu, Yu Cheng Hsu, Christian S Chan, Chaak Ming Lau, Joyce Liu, Paul Siu Fai Yip

    BACKGROUND Sentiment analysis is a significant yet difficult task in natural language processing. The linguistic peculiarities of Cantonese, including its high similarity with Standard Chinese, its grammatical and lexical uniqueness, and its colloquialism and multilingualism, make it different from other languages and pose additional challenges to sentiment analysis. Recent advances in models such as ChatGPT offer potential viable solutions. OBJECTIVE This study investigated the efficacy of GPT-3.5 and GPT-4 in Cantonese sentiment analysis in the context of web-based counseling and compared their performance with other mainstream methods, including lexicon-based methods and machine learning approaches. METHODS We analyzed transcripts from a web-based, text-based counseling service in Hong Kong, including a total of 131 individual counseling sessions and 6169 messages between counselors and help-seekers. First, a codebook was developed for human annotation. A simple prompt (“Is the sentiment of this Cantonese text positive, neutral, or negative? Respond with the sentiment label only.”) was then given to GPT-3.5 and GPT-4 to label each message’s sentiment. GPT-3.5 and GPT-4’s performance was compared with a lexicon-based method and 3 state-of-the-art models, including linear regression, support vector machines, and long short-term memory neural networks. RESULTS Our findings revealed ChatGPT’s remarkable accuracy in sentiment classification, with GPT-3.5 and GPT-4, respectively, achieving 92.1% (5682/6169) and 95.3% (5880/6169) accuracy in identifying positive, neutral, and negative sentiment, thereby outperforming the traditional lexicon-based method, which had an accuracy of 37.2% (2295/6169), and the 3 machine learning models, which had accuracies ranging from 66% (4072/6169) to 70.9% (4374/6169). CONCLUSIONS Among many text analysis techniques, ChatGPT demonstrates superior accuracy and emerges as a promising tool for Cantonese sentiment analysis. This study also highlights ChatGPT’s applicability in real-world scenarios, such as monitoring the quality of text-based counseling services and detecting message-level sentiments in vivo. The insights derived from this study pave the way for further exploration into the capabilities of ChatGPT in the context of underresourced languages and specialized domains like psychotherapy and natural language processing.

  • Open Access

    Tracking and Profiling Repeated Users Over Time in Text-Based Counseling: Longitudinal Observational Study With Hierarchical Clustering (Preprint)

    by Yucan Xu, Christian Shaunlyn Chan, Evangeline Chan, Junyou Chen, Florence Cheung, Zhongzhi Xu, Joyce Liu, Paul Siu Fai Yip

    BACKGROUND Due to their accessibility and anonymity, web-based counseling services are expanding at an unprecedented rate. One of the most prominent challenges such services face is repeated users, who represent a small fraction of total users but consume significant resources by continually returning to the system and reiterating the same narrative and issues. A deeper understanding of repeated users and tailoring interventions may help improve service efficiency and effectiveness. Previous studies on repeated users were mainly on telephone counseling, and the classification of repeated users tended to be arbitrary and failed to capture the heterogeneity in this group of users. OBJECTIVE In this study, we aimed to develop a systematic method to profile repeated users and to understand what drives their use of the service. By doing so, we aimed to provide insight and practical implications that can inform the provision of service catering to different types of users and improve service effectiveness. METHODS We extracted session data from 29,400 users from a free 24/7 web-based counseling service from 2018 to 2021. To systematically investigate the heterogeneity of repeated users, hierarchical clustering was used to classify the users based on 3 indicators of service use behaviors, including the duration of their user journey, use frequency, and intensity. We then compared the psychological profile of the identified subgroups including their suicide risks and primary concerns to gain insights into the factors driving their patterns of service use. RESULTS Three clusters of repeated users with clear psychological profiles were detected: episodic, intermittent, and persistent-intensive users. Generally, compared with one-time users, repeated users showed higher suicide risks and more complicated backgrounds, including more severe presenting issues such as suicide or self-harm, bullying, and addictive behaviors. Higher frequency and intensity of service use were also associated with elevated suicide risk levels and a higher proportion of users citing mental disorders as their primary concerns. CONCLUSIONS This study presents a systematic method of identifying and classifying repeated users in web-based counseling services. The proposed bottom-up clustering method identified 3 subgroups of repeated users with distinct service behaviors and psychological profiles. The findings can facilitate frontline personnel in delivering more efficient interventions and the proposed method can also be meaningful to a wider range of services in improving service provision, resource allocation, and service effectiveness.

  • Open Access

    A Network Approach to Understand Co-Occurrence and Relative Importance of Different Reasons for Suicide: A Territory-Wide Study Using 2002-2019 Hong Kong Coroner's Court Reports

    by Yu Cheng Hsu, Alvin Junus, Qingpeng Zhang, Clifford Wong, Florence Cheung, Joyce Liu, Ingrid D. Lui, Paul S. F. Yip

    Background: Suicide is a complex and multifaceted issue, and suicidal behaviors are often driven by multiple, interacting factors. It has been challenging to identify critical reasons for suicide using existing scientific methodologies.AIMTo identify critical reasons for suicide through novel network science methods to understand suicides and suicidal behaviors. Methods: Using data on 13,001 suicide cases with identifiable reasons obtained from the Hong Kong Coroner's Court, we modeled multiple-reason suicide as a co-occurrence network. We then analyzed the temporal and demographic changes in the structure and eigencentrality of the proposed network and conducted simulation studies based on the United Nations population projection to assess the future burden of reasons for suicide. Findings: School-related issues had the highest eigencentrality (eigencentality=0.49) for individuals younger than 20 years of age. Financial issues were crucial for adults aged 20-59 years, but their importance differed between males (eigencentality=0.51) and females (eigencentality=0.14). Physical illness (eigencentrality=0.80) was the core concern for adults over 60. years Across the Hong Kong population, the reasons for suicide appear to have shifted from financial issues in the early 2000s (eigencentrality=0.46) to issues related to physical illnesses since 2011 (eigencentrality=0.58). Simulation findings indicate that, by 2050, most suicides in Hong Kong will be due to physical illness-related issues (eigencentrality=0.69) due to the rapidly aging population. Interpretation: There have been important sex and age differences over time, in reasons for suicide. Given the predicted increasing age of the Hong Kong population over the next decades, older adults with physical illnesses appear to be the most at-risk group for suicide. This novel network analysis approach provides important data-driven information upon which to base effective proactive public health suicide prevention strategies and interventions. Funding Information: This research received no funding. Declaration of Interests: The authors declare that there is no conflict of interest. Ethics Approval Statement: Ethics approval was obtained from the Human Research Ethics Committee for Non-Clinical Faculties of the University of Hong Kong (EA210305).