Internships

Do you want to experience what it’s like to work at a high-tech scale-up? We have graduation assignments on several subjects:

On this page you’ll find more information about the graduation assignments. If you have a good idea for a graduation assignment that is not on this list, please contact us. We can discuss your ideas and maybe design a new graduation project.

8vance stimulates continuous learning. The student is asked to formulate personal learning goals to work on besides the main graduation project. Reporting, written and oral, is done in the English language.

For more information, questions or an application you can contact Anne Winters: anne.winters@8vance.com.

Graduation assignment: “Career Path Prediction”

Problem

8vance has a large set of person profiles (CVs). These profiles contain information about which jobs a person has done in his life. Based on this job history, it will be interesting to try to predict the next step in a person’s career. The data that is available consists of profiles scraped from the internet, and is therefore noisy and unstructured. If 8vance is able to predict the next career step the expectation is that this will greatly improve the matching results. As an extra challenge, an interactive way of reporting to and extracting additional info from the user can be designed.

Assignment

The project consists of several aspects. First, the available data has to be cleaned and normalized as people can describe the same job in very different ways in their profile. After that, one or more models should be trained that can predict the next job based on a list of previous jobs and potentially other information. As the data is sequential, 8vance believes that this could be modelled using an RNN architecture. 8vance has a taxonomy of function types that can serve as standardized input features. Another approach to the problem could be to use reinforcement learning or Markov chains.

The assignment should address the following parts:

  • Literature study on relevant machine/deep learning models and techniques
  • Designing new solution strategies
  • Validate these strategies by constructing prototypes

Graduation assignment: “Data Enrichment”

Problem

At this moment 8vance has a large set of job openings and personal CVs. The problem is that this data can be quite sparse i.e. missing values for features that are not extracted from natural language source material on network sites, job boards etc. One of the reasons for this is the non-standard way applicants can describe a job function. Another problem is that the missing data can be implicit, and people usually don’t bother writing it down explicitly (e.g. specific skills, competences etc.).

Usually the missing values can be found in different sources, for example a dictionary of skill-function relations evaluated by experts. Another way is to take a large set of jobs, and find common information among them. Once this common information is determined, profiles with missing data can be enhanced with this information.

Assignment

8vance would like to develop a method that can enrich a profile (job vacancy or CV) with common knowledge derived from multiple sources of truth. The student is expected to research several ways in which this common knowledge can be found/extracted and how it can be used to enrich existing profiles.

8vance believes statistical and machine/deep learning techniques can be used to improve the enrichment task. The student is expected to perform research into which techniques are most suitable.

The assignment should address the following parts:

  • Literature study on relevant machine/deep learning models and techniques, including heterogeneous ANN architectures
  • Designing new solution strategies
  • Validate these strategies by constructing prototypes

Graduation assignment: “Multi-Source Job Function Classification”

Problem

At this moment, 8vance is researching several ways to standardize function titles. 8vance has developed a function classifier that takes a title (of a vacancy, for example) in natural language, and classifies it onto a taxonomy of function types. This is largely a rule-based process. Recently, research has progressed in such a way that the actual vacancy text can be used in the classification process. The next step is to optimize the text classification model and to develop a new model that can take structured information into account (e.g. salary, work hours etc.). In addition, it would be interesting to classify other values such as required seniority and employer industry.

Assignment

The aim of the project is to develop new function classification algorithms that can take into account the different features contained in job vacancies (text, structured info etc.). In the end, the results of these algorithms should be combined into one function classification. Another important part is related to the explainability of the results; 8vance would like to know why a certain classification was made and additionally a confidence score indicating the likelihood of the classification.

The assignment should address the following parts:

  • Literature study on relevant machine/deep learning models and techniques
  • Designing new solution strategies
  • Validate these strategies by constructing prototypes

Graduation assignment: “Job/CV parsing”

Problem

In order to generate high quality feature vectors to be used in matching, 8vance would like to improve the extraction of features from job advertisements and CV profiles (parsing). At this moment the parsing is largely based on linguistic rules and is reaching its maximum performance.

8vance has a large dataset of job advertisements scraped from online job boards. This scraped data is mostly noisy (raw HTML) and has to be cleaned extensively. On the other hand, the structure of the raw HTML data could give some hints as to where relevant data can be found in the data (e.g. some websites have structured blocks with relevant information).

Assignment

The project consists of several sub-tasks in order to go from (scraped) data to clean features.

Firstly, the relevant text has to be extracted from the HTML such that it can be used in further processing or text-based machine learning algorithms or NLP techniques can be used to further extract features. This is known as ‘boiler plate removal’.

Secondly, it will be interesting to investigate whether the HTML context can provide leads to find relevant information in the data. Some websites, for example, have structured blocks with relevant information. A model can be trained to recognize such structures and extract the relevant information.

Finally, once an algorithm has been developed to remove boiler plate from scraped data, 8vance is interested in techniques that can segment and classify the remaining text into categories of interest (e.g. function description, requirements, etc.). An interesting segmentation method to be researched might be to use a ‘shortcut’ and convert the HTML into an image and subsequently perform segment detection on the image (and OCR the result).

8vance believes that a combination of Machine/Deep learning techniques and smart algorithm design can accomplish the above-mentioned tasks and can result in a product that can parse the raw HTML data into useful features and does so with higher quality than the currently used methods.

The assignment should address the following parts:

  • Literature study on relevant machine/deep learning models
  • Designing new solution strategies
  • Validate these strategies by constructing prototypes

Graduation assignment: “General Model for User Feedback”

Problem

8vance would like to better incorporate user feedback into its matching process. The idea is that users can like or dislike a given match, and in this way influence the system to generate better matches. 8vance currently has such a feedback system in place for its candidate matching solution. The challenge in this project is to improve upon the current system, and also to develop a more general approach that can be implemented across all 8VMT verticals. 8vance has a set of user feedback data available.

Assignment

The current approach is focused on adjusting/regenerating user search-profiles based on the likes and dislikes of a user. One could think of more sophisticated models of incorporating user feedback, such as collaborative filtering where also the preferences of other users are taken into account. One could even think of reinforcement learning approaches.  Part of the assignment is to find innovative ways to extract feedback from users (e.g. chatbot). Also bootstrapping techniques should be considered to continuously improve the matching-feedback loop.

The assignment should address the following parts:

  • Literature study on relevant machine/deep learning models and techniques
  • Designing new solution strategies
  • Validate these strategies by constructing prototypes