Eval: Advanced emotion analysis for complex scenarios based on a Ph.D. dissertation

[tuanlemau]

Thank you for contributing an eval! ♥️

🚨 Please make sure your PR follows these guidelines, failure to follow the guidelines below will result in the PR being closed automatically. Note that even if the criteria are met, that does not guarantee the PR will be merged nor GPT-4 access granted. 🚨

PLEASE READ THIS:

In order for a PR to be merged, it must fail on GPT-4. We are aware that right now, users do not have access, so you will not be able to tell if the eval fails or not. Please run your eval with GPT-3.5-Turbo, but keep in mind as we run the eval, if GPT-4 gets higher than 90% on the eval, we will likely reject since GPT-4 is already capable of completing the task.

We plan to roll out a way for users submitting evals to see the eval performance on GPT-4 soon. Stay tuned! Until then, you will not be able to see the eval performance on GPT-4. Starting April 10, the minimum eval count is 15 samples, we hope this makes it easier to create and contribute evals.

Eval details 📑

Eval name

complex-emotion-analysis

Eval description

This eval aims to evaluate GPT’s ability to analyze emotions from a complex scenario, one that may elicit more than just one emotion, although only one emotion is dominant. The scenarios have been evaluated by 40 human subjects in a study as part of my Ph.D. dissertation at the Massachusetts Institute of Technology (MIT) and also published in Nature Communications (https://www.nature.com/articles/s41467-021-25352-6 ). In this eval, 30 scenarios have been selected from the set of 604 scenarios in the study such that GPT-3.5 only achieve 23% accuracy in predicting the emotion category that most human participants indicated best reflect the emotion experienced in a given scenario. Scientific Reference:

1. Le Mau, T., Hoemann, K., Lyons, S.H. et al. Professional actors demonstrate variability, not stereotypical expressions, when portraying emotional states in photographs. Nat Commun 12, 5037 (2021). https://doi.org/10.1038/s41467-021-25352-6
2. Le Mau, T, Toward Understanding Facial Movements in Real Life. MIT Library. Ph.D. Dissertation.

What makes this a useful eval?

This eval provides a state-of-the-art evaluation of the performance of an advanced emotion analyzer in a given complex text (scenario). Firstly, to my knowledge, the dataset presented in my Ph.D. dissertation is unique in both the emotional complexity of the scenarios. The person in each scenarios was rated to may have experienced more than one emotions. The scenarios in this evaluation have been chosen so that one emotion is predominant but other emotions are present as well. Secondly, the evaluation data is unique in the rating of the scenarios. Typical datasets for emotion analysis only use about 2, at most 3 raters. Each of the scenarios provided in this eval set were evaluated by 40 human raters. Moreover, human raters selected from a much wider range of emotions (13 categories) instead of just the typical 5 categories of Joy, Sadness, Anger, Fear and Surprise. Therefore, this evaluation dataset can help GPT and other generative models to achieve much higher accuracy in emotion analysis of real-life, emotionally complex scenarios (which tend to happen much more often than simple ones). Any generative models able to get high accuracy (> 80%) from this evaluation dataset will be the new state-of-the-art model in textual emotion analysis.

Criteria for a good eval ✅

Below are some of the criteria we look for in a good eval. In general, we are seeking cases where the model does not do a good job despite being capable of generating a good response (note that there are some things large language models cannot do, so those would not make good evals).

Your eval should be:

• [x] Thematically consistent: The eval should be thematically consistent. We'd like to see a number of prompts all demonstrating some particular failure mode. For example, we can create an eval on cases where the model fails to reason about the physical world.
• [x] Contains failures where a human can do the task, but either GPT-4 or GPT-3.5-Turbo could not.
• [x] Includes good signal around what is the right behavior. This means either a correct answer for Basic evals or the Fact Model-graded eval, or an exhaustive rubric for evaluating answers for the Criteria Model-graded eval.
• [x] Include at least 15 high quality examples.

If there is anything else that makes your eval worth including, please document it below.

Unique eval value

Each scenario in the eval set was rated by 40 human raters (typically 2 or 3) and across 13 emotion categories (typically 5). The dataset can improve the accuracy of generative models in analyzing real-life emotionally complex scenarios.

Eval structure 🏗️

Your eval should

• [x] Check that your data is in evals/registry/data/{name}
• [x] Check that your yaml is registered at evals/registry/evals/{name}.yaml
• [x] Ensure you have the right to use the data you submit via this eval

(For now, we will only be approving evals that use one of the existing eval classes. You may still write custom eval classes for your own cases, and we may consider merging them in the future.)

Final checklist 👀

Submission agreement

By contributing to Evals, you are agreeing to make your evaluation logic and data under the same MIT license as this repository. You must have adequate rights to upload any data used in an Eval. OpenAI reserves the right to use this data in future service improvements to our product. Contributions to OpenAI Evals will be subject to our usual Usage Policies (https://platform.openai.com/docs/usage-policies).

• [x] I agree that my submission will be made available under an MIT license and complies with OpenAI's usage policies.

Email address validation

If your submission is accepted, we will be granting GPT-4 access to a limited number of contributors. Access will be given to the email address associated with the merged pull request.

• [x] I acknowledge that GPT-4 access will only be granted, if applicable, to the email address used for my merged pull request.

Limited availability acknowledgement

We know that you might be excited to contribute to OpenAI's mission, help improve our models, and gain access to GPT-4. However, due to the requirements mentioned above and high volume of submissions, we will not be able to accept all submissions and thus not grant everyone who opens a PR GPT-4 access. We know this is disappointing, but we hope to set the right expectation before you open this PR.

• [x] I understand that opening a PR, even if it meets the requirements above, does not guarantee the PR will be merged nor GPT-4 access granted.

Submit eval

• [x] I have filled out all required fields in the evals PR form
• [ ] (Ignore if not submitting code) I have run pip install pre-commit; pre-commit install and have verified that black, isort, and autoflake are running when I commit and push

Failure to fill out all required fields will result in the PR being closed.

Eval JSON data

Since we are using Git LFS, we are asking eval submitters to add in as many Eval Samples (at least 5) from their contribution here:

View evals in JSON

Eval

{"input":[{"role":"system","content":"As an emotion analyzer, your job is to identify the predominant emotion the person in the sentence experience and respond with the corresponding one-word emotion category. The available emotion categories are: Amusement, Anger, Awe, Contempt, Disgust, Embarrassment, Fear, Happiness, Interest, Pride, Sadness, Surprise, and Shame."},{"role":"user","content":"He is a highway patrolman peering into a car filled with smirking teenagers."}],"ideal":"Interest"}
{"input":[{"role":"system","content":"As an emotion analyzer, your job is to identify the predominant emotion the person in the sentence experience and respond with the corresponding one-word emotion category. The available emotion categories are: Amusement, Anger, Awe, Contempt, Disgust, Embarrassment, Fear, Happiness, Interest, Pride, Sadness, Surprise, and Shame."},{"role":"user","content":"She's high on ecstasy, She's jumped up on stage to hug the lead singer of her favorite band, and she can't figure out why a couple of security guards are dragging her off."}],"ideal":"Surprise"}
{"input":[{"role":"system","content":"As an emotion analyzer, your job is to identify the predominant emotion the person in the sentence experience and respond with the corresponding one-word emotion category. The available emotion categories are: Amusement, Anger, Awe, Contempt, Disgust, Embarrassment, Fear, Happiness, Interest, Pride, Sadness, Surprise, and Shame."},{"role":"user","content":"He's a powerful Hollywood producer coming on to a young, vulnerable, na\u00efve ingenue."}],"ideal":"Interest"}
{"input":[{"role":"system","content":"As an emotion analyzer, your job is to identify the predominant emotion the person in the sentence experience and respond with the corresponding one-word emotion category. The available emotion categories are: Amusement, Anger, Awe, Contempt, Disgust, Embarrassment, Fear, Happiness, Interest, Pride, Sadness, Surprise, and Shame."},{"role":"user","content":"It's Halloween, and he is greeting a bunch of little trick-or-treaters at the door in a way he figures they'll never forget."}],"ideal":"Amusement"}
{"input":[{"role":"system","content":"As an emotion analyzer, your job is to identify the predominant emotion the person in the sentence experience and respond with the corresponding one-word emotion category. The available emotion categories are: Amusement, Anger, Awe, Contempt, Disgust, Embarrassment, Fear, Happiness, Interest, Pride, Sadness, Surprise, and Shame."},{"role":"user","content":"He's a mean older brother making fun of his sister for crying after he pushed her off her bike."}],"ideal":"Amusement"}