Project
Please read this page fully and carefully. It contains the necessary guidelines for projects in CS886 (Affective Computing)
General Information
-
Students must complete an individual course project. Small groups (2-3 people) must be approved first, and are expected to be proportionally more involved, and must have a clear delineation of work responsibilities amongst the group members. One clear reason to form a group is in the case of a project requiring ethical approval see below. One student lead this process while the other(s) develop, etc.
-
The project must be an implemented working system, a user study, a data analysis or replication of a previous result, or a new model or theory. Literature reviews may be accepted, but must have a clear analytical component (discuss with the instructor if you plan to do this).
- A written project proposal (in PDF ONLY) must be submitted on LEARN before end of day October 7th, 2024 (anywhere on earth). However, if you are planning a project involving human subjects, you need to start much earlier - see below under Ethical Approval. Project proposals should be about a page long and include 5-10 references. Formatting is freeform, but describe the purpose and motivation of your idea, as well as a basic outline of how you plan to approach it.
- A written report must be submitted on or before last day of class 2024 (anytime up to the end of the day anywhere on earth.) Reports are also submitted on LEARN.
- Projects must be 8 pages long, and formatted according to the IEEE Conference (two-column) Template.
- Projects must be submitted to the appropriate dropbox on LEARN as a single PDF ONLY entitled <STUDENTNAME>_<STUDENTNUMBER>.pdf before the due date.
- Materials submitted in anything other than PDF, or with any incorrect formatting, or that contravene any of the integrity and citation guidelines below, will be rejected without evaluation.
Ethical Approval
Important: If your plan to involve human participants in your project (e.g. do a survey, test an instrument), you
must get ethical approval, and you must start this process
immediately as it can take six weeks or more. See
Office of Research Ethics and in particular
this page for details. I don't have a blanket ethics procedure for the course, as projects are too varied to allow for that, so each student or group is responsible for their own ethics applications if needed.
Evaluation
The project reports will be evaluated on three main criteria, with weights as shown.
These are used as guidelines for the instructor when evaluating the work.
- Completeness (50%): does the report state contributions, claims, assumptions, strengths and weaknesses? Does the report have an abstract that clearly defines the work and results? Does the report give a complete overview of the methods used? Does the report provide a discussion that appropriately summarises the work? Are there experimental or theoretical results to back up the claims? Is there a conclusion that adequately summarises the work, including the main claims and results? Are there sufficient references?
- Clarity (30%): is the report clear, readable, and free of spelling and grammar errors, and presented using the required format? If Generative AI is used, it is properly declared?
- References (10%): are the references correctly formatted, complete (e.g. including page numbers, book titles, years, etc), and are the guidelines for citing wikipedia and other online content respected (see below)?
- Originality (10%): does the report uncover something novel? You do not need to invent time-travel to get this extra 10%! An original take on an old problem, or an original problem statement can be sufficient. Clearly state in your report what is novel about your work!
Academic Integrity
All submitted work (assignments and project reports) should be the submitting student's own work, and should bear his/her name as author.
Any evidence of writing from a different person or source will be considered plagiarism and will be dealt with as such (see University policy below). A few simple rules may help you here:
- ALWAYS write your own submitted work. (see note on generative AI use below)
- CLEARLY indicate contributions from anyone else: enclose the text in "quotations" and CLEARLY indicate the source right next to the text. Just citing the work that you are quoting from at the end in the bibliography is NOT sufficient. Rule 2 must be applied IMMEDIATELY upon putting any text or image that is not your own into your document. Don't wait, you'll regret it later when you submit your work but have forgotten to go back and put in the citation.
- DON'T cut and paste.
- If you do cut and paste (remember: don't cut and paste - see rule 3), then apply rule 2 IMMEDIATELY.
- DON'T cite Wikipedia or other publicly generated web content. Read about citing Wikipedia or other web content.
Citations
Citations are an important part of any scientific writing. They put your work in context, and show its relationship to other work. A big part of the power of a citation comes from its trustworthiness: the source is such that you have good reason to believe that a group of professional scientists has read and evaluated the paper, and so it contains mostly valid scientific results (although they may have made a mistake, but then it is up to you to find it!).
If you are talking about someone else's work, then you put a reference (properly formatted) to the work in a bibliographic section at the end of your paper, and cite the reference after you have discussed the work:
Carroll's use of poetry is an important part of his work, as it contextualizes and motivates the compositional aspects of the work (Carroll, 1871).
However, if you are directly quoting someone else's work (do this very sparingly), then you need to enclose the text you are quoting in quotation marks, and put the reference, with a page number,
immediately after the text you are quoting.
The poem opens with the line: "The sun was shining on the sea" (Carroll, 1871, p.72).
And, then, at the end of the document, you list the full reference.
Carroll, Lewis. Through the Looking Glass and What Alice Found There. MacMillan, 1871.
The actual format of the references will change depending on the journal or conference you are writing for. If you are using LaTeX, then you can simply change the bibliography style, see
here for a guide to LaTeX sytles. If you're using Word™, then I question your sanity but nevertheless wish you the best of luck.
Wikipedia (or other publicly generated web content) is not a trustworthy source of information. Wikipedia entries can be generated by anyone, and may be left un-checked for months or years. Any references to Wikipedia in your submitted coursework for CS886 will not be considered valid, and significant marks will be deducted. However, you may use Wikipedia to seek the original references, and cite those instead.
The only exception to the "don't cite Wikipedia" rule is if the content you are citing is specifically about Wikipedia itself. e.g. "In my research on the trustworthiness of Wikipedia, I found that there was a significant error on the page about affective computing (reference to webpage and date of access)". In such cases, you must include the date that you accessed the content on and the exact URL.
Citations to works that are online only (not in print) are perfectly fine, but must include a journal name, volume number, and other relevant citation material, and/or a digital object identifier (DOI). You can also cite a web page from a well-established organisation, but you should include as much reference material as possible, e.g. an ISBN number or DOI.
Wikipedia itself has a good page about citing Wikipedia.
Generative AI
All usage of Generative AI (e.g. ChatGPT, etc) is permitted, but must be declared on all submitted work. Any work using GPT-like tools without declaring will be given zero marks. Also, note that GPT-like tools will generate the uniform content across users. Submitted work will be graded in part on originality, so any work using GPT-like tools may not obtain these marks.
Project ideas
- Develop a probabilistic and/or decision theoretic version of the OCC model of emotions. There have been some efforts towards this goal, but none have made a convincingly clear and well-grounded model. See the papers by Reisenzein as a starting point.
- Develop an implementable model of the Theory of Social Commitments (Lawler Thye and Yoon - see readings). This could be done using BayesACT. Talk to me if interested.
- Develop a method for adding emotions to your favorite video game. Present a model describing precisely how this could be done, with some simulation results using a "toy" version of your video game.
- Investigate moral decision making using affective models, such as BayesACT. Trolley problems are a good place to start. See the paper by Wilson and Scheutz. Others on readings page.
- Write an app to play the Prisoner's Dilemma (or other social dilemma) with humans, using facial expressions of emotion and the Bayesact model (or other model of emotions). See the papers by Antos and/or Stratou under readings/dilemmas.
- Model a negotiation task using a theory of affect. For example, in Chapter 8 of Social Commitments in a Depersonalized World (Lawler, Thye and Yoon - ask me for a copy), a social inequality experiment is described where participants play a "split the pot" game. Beliefs about social power come into play resulting in highly unequal payoffs, but not unequal satisfaction with the outcomes. See also the paper by DeMelo, Carnevale and Gratch under readings/dilemmas.
- Develop an emotional speech recognition system. Based on audio signals from the environment, generate E,P,A ratings of things the user has said, or other sounds that are heard.
- Develop an animated avatar "face" that will communicate affective expressions to the person washing their hands.
- Sentiment Analysis of Text. Look into sentiment analysis and applications to text. One idea is to look at children's books and attempt to do an analysis of plot/story-lines, e.g. using affect control theory or appraisal theories. Children's books might be a good place to start because the plot is usually very simple and the language is easy to parse. Frog and Toad by Arnold Lobel is a good example. You could also look at more complex examples, even socio-political texts, to try to gain a deeper understanding of current social issues like terrorism or racism.
- Emotional chatbot
Create a chatbot that uses some emotional reasoning to respond to what a user says. You can use Bayesact or Interact to do the emotional reasoning part, and all you need is to provide:
- a method for mapping chats (of the user) to E,P,A values
- a method for mapping E,P,A values back to chats (for the chatbot)
You may find word net or Sentiwordnet useful for mapping strings/words to/from EPA values.
If you don't want to use affect control theory, you can come up with your own hand-crafted set of appraisal rules, etc. The use of emoticons (e.g. :), :0...) can be useful in this regard.
- Cyber-bullying and Cyber-predation
This project will work on building a cyber-bullying and/or cyber-predation "detector". The internet is a dangerous place for many, especially those in vulnerable positions (e.g. young, old, or with a cognitive disability). For example, teenagers suffer from bullying by peers on social networks like Facebook and Twitter. Children can be targeted by sexual predators, and the elderly can be targeted by financial predators. This project will look at how to detect bullying behaviours (an emotional action) online, and/or how to detect people posing as a different identity than their own self. For exampe, a bully will try to build a position of power over a target, and this could be detected by analysing tweets or facebook posts. A sexual predator may "pose" or pretend to be someone that he is not (e.g. a older male posing as a teenage boy and "grooming" a teenage victim so that they perform sexually). This could be detected by comparing posts from the "posers" to posts of the "posed" identity, and flagging up behaviours that seem suspicious.
- Emotions in Bayesact
One thing we have not experimented with very much is emotions in Bayesact. The equations are implemented but much work needs to be done to verify and investigate them. Your project will be to verify them in simulations or real experiments.
- Affective Machine Translation
Build a machine translation system that uses affect to disambiguate words.
- Comparing Bayesact and OCC/Appraisal models
In class, we did a brief analysis of Steamboat Willie using the OCC model of emotions. This project will be an empirical study to test the idea that many appraisal rules are implicitly encoded in (Bayes)ACT. Similarly to applying the OCC rules to movies/books, this project will be to label the scenes in ACT's subject-verb-object format to see which emotions INTERACT or BayesACT comes up with. The hypothesis is that there will be a huge overlap, which would suggest that it may be possible to get rid of many explicit appraisal rules, as they kind of emerge from the more general (Bayes)ACT algorithm.
- Affect Control Theory and the Prisoner's Dilemma
The Prisoner's Dilemma is an abstract representation of a cooperative situation in which two (or more) agents can achieve higher global good by acting in concert than by being selfish. Humans are consistently able to achieve the optimal solution by acting non-rationally (individualistic rationality). It is thought that emotional reasoning can play an important role in ensuring this happens. This project will be to investigate emotions in the prisoner's dilemma. Start by reading the papers by Maarten Wubben (see papers on the readings page under decision-making)
- Planning methods for POMDPs modeling emotion
Look into planning methods for continuous action POMDPs (talk to me about this)
- Look at the papers in the conference proceedings, find one that interests you and re-implement the model or experiment.