Abstract B013: Emotional tone classification as a tool for psychosocial risk detection in oncology: An AI-powered chatbot
Bibliographic record
Abstract
Abstract Introduction: Emotional distress is prevalent among cancer patients undergoing systemic therapy, yet it remains significantly underreported and undertreated due to limited clinical time, stigma, and inadequate detection tools. MARVINA is an AI-powered chatbot designed to facilitate symptom management for breast cancer patients. We aim to incorporate psychosocial risk detection features into the chatbot to help identify patients in need of additional psychosocial support or intervention, and to deliver tailored resources accordingly. To explore this functionality, we developed a module to perform emotional tone classification of patient text input. Methods: We cleaned an annotated corpus of 3,750 messages drawn from a Mendeley dataset of patient and caregiver posts across five cancer types (brain, colon, liver, leukemia, lung), sourced from Reddit, DailyStrength, and HealthBoards. We categorized posts with a four-item emotional tone scale: very negative (n=1,000), negative (n=1,000), neutral (n=1,000), and positive (n=750). Using a 70/15/15 training/validation/testing split, we fine-tuned a pre-trained DistillBERT classification model—originally trained on a five-item scale (very negative to very positive)—while applying class weights to address label imbalance. To simulate the informal structure of real-world conversations and deepen comprehension, we expanded the model’s vocabulary to include emojis, newlines, emails, URL links, and resized the embedding layer accordingly. Recall, precision, accuracy, and a confidence score—derived from the softmax output of the model’s final layer—were computed for each prediction. Results: Evaluations suggest strong model performance in distinguishing emotional tone, with particularly high precision for very negative content and high recall for neutral content. Metrics for very negative (89%/86%), negative (79%/86%), neutral (91%/84%), and positive (82%/85%) are presented in (precision/recall) format. The model attained an overall accuracy of 85% across all classes. For example, the input “I’m feeling much better today after the treatment 😊. Hope everything will go well in the future!” was classified as positive with 90% confidence. The input, “I'm exhausted. The chemo is unbearable, and every day feels worse than the last. I’ve lost my hair, my strength, and honestly, my hope.” was classified as very negative with 83% confidence. Conclusions: Our module represents an innovative step toward leveraging AI for emotional tone classification in oncology care, laying the groundwork for enhanced mental health monitoring and risk detection within chatbots, while supporting patient self-management in cancer care. Future directions include exploring psychosocial risk evaluation by implementing tailored chatbot responses and automated alerts to care teams in response to very negative or recurrently negative inputs. Conversely, messages that promote resilience and self-efficacy could be used to reinforce a positive tone. Citation Format: Nissim Maxim. Frija-Gruman, Sebastian Villanueva, Yuanchao Ma, Esli Osmanlliu, Sylvie Lambert, Marie-Pascale Pomey, Tarek Hallal, David Lessard, Kim Engler, Jia Lin, Jamil Asselah, Bertrand Lebouché. Emotional tone classification as a tool for psychosocial risk detection in oncology: An AI-powered chatbot [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Artificial Intelligence and Machine Learning; 2025 Jul 10-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(13_Suppl):Abstract nr B013.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.009 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.013 | 0.007 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".