Ready-To-Go And Customized Prediction Models
White Paper Series Part 6
At Vima, we develop advanced behaviour modelling and artificial techniques to automate various screening processes of individuals based on their video recordings. Across application domains, our automated systems provide reliable predictions of an individual’s personality traits, emotional state, and soft skills focusing on social, communicative and professional competencies.
Based on our existing databases of professionally active individuals from different backgrounds, Vima has developed robust and general models that are able to predict several traits, skills and states with good overall reliability and accuracy.
Vima’s general models can be used as-is (ready-to-go). In addition, given that every job and every company is unique, Vima also develops new models and finetunes existing models to the specific job and client (company) demands in order to improve case-specific accuracy (e.g., person-job match, specific reference group from within a company or across an industry). This additional layer of customization can be modelled based on sufficient data provided by the client.
Two streams of input are required to customize a model to a specific target job or company.
1. Mapping the Job or Company Profile
Together with the client company, Vima’s behavioural scientist answers a number of questions that aim to operationalize relevant constructs: What skills and personality profile does the ideal employee or job candidate have (job fit)? What are the team dynamics (team fit)? What are the company’s values and norms (culture fit)?
2. Mapping the Candidates’ or current Employees’ Profiles:
Vima continues to enlarge its proprietary database of thousands of videos of individuals with different cultural backgrounds, gender, language, age, and professional status. Population-specific models are built from these extended data collections with increased accuracy and representativeness to the tested individuals.
To train a customised company or job-specific model, Vima requires a sufficient sample of video recordings of individuals who are either applying to a particular job or who are currently employed and working under the relevant job function.
The videos depict each individual candidate or employee in a standardised recording format (in the case of existing data, visual and audio qualities are verified by Vima). He or she is asked to answer a few questions whilst being recorded, much like a short self-presentation and mini-job interview of 2-3 minutes. This process can take place within the customer’s environment or hosted by Vima’s software, complying with GDPR and Swiss Data Privacy law.
These videos are then processed through two independent pipelines.
- A pool of human experts in behavioural assessment evaluates the job candidates or current employees as defined in step 1 (mapping the job profile). State-of-the-art multimodal behaviour detection algorithms (face, voice, body) process the same subjects to automatically extract a wide range of behaviour cues.
- Advanced machine learning techniques are then applied to both annotation data and video data (including meta-data about the recorded individual such as job function) to train a custom model that predicts the expert evaluations. If current employees are screened, a concurrent validation procedure can immediately be started by feeding the model with individual performance indicators. If actual job candidates are being screened, job success from hired candidates can be assessed after a certain time-lapse (6- 12 months), enabling the model to fine-tune to the company’s actual needs.
We run strict scientific testing on any custom prediction model to measure performance in terms of reliability and validity. This data- and behaviour-driven method renders Vima’s personal assessment transparent and unaffected by subjective differences in recruiters (such as biases or stereotypes during the job interview) or in candidates (self-report bias).
An important advantage of customized models follows from the richness of the data provided by the company: it allows to expand on the assessment by integration with other relevant sources of information (e.g. performance indicators, company profile) and in a longitudinal framework (long-term prediction of job success). This is how multiple custom norm groups are built and adapted, to which new data points are compared.
P1. Competent: The person has the necessary ability to do something successfully.
P2. Hard-working: The person tends to work with energy and commitment; is diligent.
P3. Professional: The person is competent, skilful, or assured.
P4. Motivated: The person is enthusiastic and determined to achieve success.
P5. Resilient to stress: The person is resilient, able to adapt to stress.
Personality Traits (“Big 5”)
– Open to new experiences, complex
– Conventional, uncreative
– Dependable, self-disciplined
– Disorganized, careless
– Extraverted, enthusiastic
– Reserved, quiet
– Critical, quarrelsome
– Sympathetic, warm
5. Neuroticism aka emotional stability
– Anxious, easily upset
– Calm, emotionally stable
Conclusion of the White Paper Series
These scientifically oriented series introduced the field of Behavioural Intelligence. Human behaviour can best be understood by integrating social and behavioural sciences with computational science and technology. Vima is situated at this intersection, offering validated tools for personality assessment using behavioural AI.
We introduced core concepts inherited by psychology and explained how internal traits and states can be measured using standardized behaviour-based observation. Vima’s technology infers internal traits and states by capturing the behaviours that human experts consistently perceive and use in their assessment. This approach remedies some of the challenges faced by traditional methods such as self-report questionnaires. In the end, you are what you do, not what you say you do.
Vima cultivates high-quality behaviour-based assessment that is scalable by developing machine learning models that automate behaviour analysis and inference of emotional states, personality traits and soft skills as perceived by others. Given that professional decision making can be biased, we revealed how Vima limits bias through every step in its technology, thus meeting the highest standards for building quality AI.
Finally, it does not stop here. In addition to Vima’s ready-to-go prediction models, tailored models can be customized for a specific company or industry profiles and for job mapping. Vima’s behavioural AI solution for person assessment can then be fully integrated with a company’s customs and practices.
In sum, the current surge in video recording of professional interactions (e.g. video CV’s, webinars, video conferences, sales pitches) represents a true opportunity to develop automated systems for behaviour-based inference of personality traits, soft skills and emotional states. This approach complements existing person assessment methods, which are often biased and difficult to scale in a cost-effective way. Vima’s behavioural intelligence solutions leverage fast and scalable algorithms built upon ground truth data that is both large in size and high in quality.
Vima continues to pave the way forward in translating scientific knowledge and know-how from behavioural intelligence to usable business solutions in-person assessment.
List of major soft skills and personality traits selected and implemented in VIMA’s technology:
- Recurring in the professional practice of human assessment
- Rcientifically validated by published research
C1. Clear: The message conveyed was easy to perceive, understand, or interpret.
C2. Concise: The person was giving la lot of information clearly in a few words.
C3. Communicative: The person was willing, eager, or able to talk or impart information.
C4. Persuasive: The person was good at persuading someone to do or believe something.
C5. Confident: The person showed confidence in conveying the Message.
S1. Sociable: The person is willing to talk and engage in activities with other people.
S2. Positive: The person is constructive, optimistic, or confident in activities with other people.
S3. Friendly: The person is kind and pleasant to interact with.
S4. Enthusiastic: The person has intense and eager enjoyment or interest in activities with other people.
S5. Good at teamwork: The person is collaborative, able to work with others.
- In charge, takes control
Shows leadership, takes responsibility
- In charge, takes control
- Positive overall impression