International students to USA universities from the beginning of their application submission are at a disadvantage with competition for seats that grows every year and may even be different from one term to another depending on the applicant pool. However, the biased practices don't stop there but they begin from the screening process. International students are expected stronger backgrounds, better GPA compared to local students, better essays where their english is evaluated more stringently even a grammar or spelling mistake is an issue, and many go through an FBI database profile check. Yes, there is racial profiling involved here. And, if the program is affiliated with defense or military then one can be sure to be scrutinized for religious affiliations which becomes apparent based on the application profile from such things as place of birth, name, and from essays. First amendment constitution rights don't hold any weight on an international student application screening process and affirmative action is only for namesake. And, no controls on the application for diversity, equity, and inclusion. One thing to note is that on most USA institutions the research assistants usually have very similar backgrounds. They are either locals, europeans, or asians with predominately alternative religious or non-religious affiliations. In public universities they are required to comply with anti-discrimination laws. However, in private universities this may not be the case. And, often these laws are abused by individuals under the covers of paperwork and cultural biases where it is extremely difficult to apply transparency and accountability. In fact, many admissions processes don't even provide any feedback to applicants. The academic admissions process is very much subjective. This is equally why universities need to resolve biases by introducing AI into the system of screening applications. And, not only accepting or rejecting applications but providing the right sense of constructive feedback that can help them in their future endeavors whether that be a re-application in the future, an acceptance at the university, or beyond. The following are some suggestions on how an application could be evaluated via AI:
- Key/Value extractions in JSON form from applications, then store in a database
- Key/Value extractions in JSON form from resume, then store in a database
- Key/Value extraction in JSON form from academic transcript, then store in database
- Automated essay scoring using NLP methods
- Automated resume scoring using NLP methods
- Basic distribution curves of GPA in pool of applicants
- Basic distribution curves of standardized test scores of applicants
- Identify outlier class attributes using unsupervised clustering methods
- Building a bayesian model for uncertainty reasoning for causal inference, this could be in the form of a factor graph to identify whether this applicant will :
- struggle to maintain a 3.0 GPA,
- whether they are likely to accept if given an offer,
- whether resume experiences can be considered in lieu of lower GPA,
- the degree of course rigor for which they attained a high GPA,
- the overlap of courses in the alternative degree attained,
- the likelihood of meeting the prerequisites,
- whether they are likely to dropout part way through the course,
- whether they were top of their class in their respective peer group given the percentile,
- the degree of their intents, interests, experience, goals, maturity, and peers match to the program
- Produce a Knowledge Graph representation of the applicant profile that could be queried for 5W1H question/answering.
- Identify the key classification classes that every applicant must have to build a diversified pool of offers - this could be a combination of supervised regression and unsupervised clustering methods
- A graph-machine learning based recommendations system that ranks the candidates in order of priority taking into account outliers
- Build an affective computing model using symbolic reasoning to identify intents and interests then feed that back into the recommendations process
- Apply a fraud detection mechanism to identify fake application documents
- Apply a lie detection mechanism to identify whether the applicant actually did these extracurricular activities like volunteering, whether they actually did face such hardships, or whether they did have these work experiences
- Apply a financial evaluation of applicants to see whether they will be able to financially cope with the tuition and fee payments or will they struggle significantly, then look to evaluate recommendations for financial aid if they meet such criteria.
- Build a feedback loop mechanism to enhance the application structure in the way they are worded to analyze for biases and build an explainability model
- Build a criteria model for fairness then apply such transparency and accountability measures for errors in the feedback loop
- Build a constructive feedback loop for applicants identifying the class attributes that held them back and what they could do to improve their chances of future acceptance, although this may vary based on any given applicant pool for the term, at least it could be evaluated from historical trends. Most rejection letters are worded in a standard form which is unhelpful to applicants.
- For Phd applicants this may also have a separate step to evaluate the strengths of the department vs the interests of the applicant and whether an appropriate advisor could be provided.
- References could further be assessed with NLP methods of extraction, and as a feedback loop into the recommendations, question/answering, causal inference, financial evaluation, fraud, deception, quantifying biases, and other forms of assessment criteria
- Identify similarity trends between applicants using nearest-neighbor methods
- Identify regression method for whether applicants meet the relevant prerequisites
- Build a chatbot that assists applicants through the pre-application and post-application process. The objective should be of converting the applicant into a potential student and treating the individual as a customer. This chatbot should then take into account customer relationship management and affective computing for sentiments and emotions.
- Anonymization and masking of potential key attributes that could pose as an underlining bias in the decision like place of birth or name.
- Localization of applications if a threshold needs to be maintained between international, in-state, and out-of-state applicant offers.
- If there are publications they could be checked for citation scores and validated for theory correctness and coverage
- If there are any awards won they could be verified and validated
- Identify a key set of behavorial attributes that may be needed for a successful student on the course then measure against those set targets
- Making sure that correlation does not imply causation
- Convert non-refundable fees into refundable retainer fees so when an applicant is declined the application fee is refunded. Why penalize applicants that got declined, in fact it will only push more applicants to apply that would boost the ranking of the university. Why get applicants to apply if it is obvious they will be rejected at the expense of boosting university rankings. Only accept application fees from those that you have accepted to the university and let the fee go someway towards the tuition and fees.
