Abstract Body

Background

Identifying HIV prevention needs is challenging. Common approaches (e.g., public health campaigns, HIV risk scores) have limited sensitivity. Sentiment analysis uses natural language processing, text analysis, and computational linguistics to identify and study affective states and subjective information. This study explored a novel use of sentiment analysis with WhatsApp messages to identify unrecognized HIV prevention needs among young women in Kenya.

Methods

Prior work established key ethical protections (e.g., privacy, confidentiality). Young women (18-24 years) who used WhatsApp were enrolled from 4 clinical sites in Kisumu, Kenya. After completing a socio-demographic questionnaire, participants securely transferred all WhatsApp messages from the prior 6 months and provided anonymized relationship labels (e.g., friend, sexual partner) for their 20 most frequent contacts. Data were de-identified, and all incoming messages and photos were deleted. WhatsApp data were pre-processed using a downloaded large language model to translate Kiswahili and Dholuo to English and account for slang linguistic patterns. We explored multiple approaches to topic modelling in the WhatsApp data using short text clustering and context relevant seeds as well as textual counts. We also analyzed topics according to known contacts. We used Pearson correlation to compare topics with known HIV risk scores (e.g., VOICE) and uniform manifold approximation and projection to assess data dimensions as vectors.

Results

Of 413 eligible young women, 13 declined to provide WhatsApp data. Participants had a mean age of 22 years and a VOICE risk score of 6 (range 2-9); 16 had insufficient data for analysis. The remaining 386 women provided a mean of 25 chats with unique individuals. Per the Figure, topic models had differential correlation with the VOICE risk score; dimensional analysis identified discrete populations with relation to VOICE risk scores. Messaging a sexual partner the most was also associated with a higher VOICE risk score (median 7 vs 5 for a friend).

Conclusions

This study provides proof-of-concept that sentiment analysis of WhatsApp messages can yield meaningful individual-level information about HIV prevention needs based on routine communication patterns. Future work will compare sentiment analysis findings with HIV incidence to optimally determine its ability to identify unrecognized HIV prevention needs, thus potentially expanding HIV care to those not currently engaged in care.

Table or Figure