Consultancy – Natural Language Processing (NLP) Researcher, Accelerator Lab Network

Background

The UNDP Accelerator Labs Network is the largest learning network in the world. It includes 91 public innovation labs and a global curation team, all embedded within UNDP’s global architecture and country platforms.

Each lab surfaces and reinforces locally sourced solutions to development challenges, and mobilizes a wide and dynamic range of partners who contribute knowledge, resources, and experience. The goal is to transform traditional development approaches by introducing new protocols, backed by evidence and practice, to accelerate the testing and dissemination of solutions within, and across countries, while at the same time enabling the global community to collectively learn from local knowledge and ingenuity at a speed and scale that our societies and planet require.

To support these learning outcomes, the UNDP Accelerator Labs global team is building a Network Learning Strategy and Prototype for knowledge management. The strategy has a strong digital component, that builds on the idea of “going where the information is”. The goal is to facilitate the detection of “where” knowledge is held in the network, and “why” it can be assumed to be held there, rather than trying to capture, infer, or extrapolate precisely “what” that knowledge is. This requires “listening” to the activities of the network, by centralizing and making sense of the content the labs put out, as well as the digital traces they leave behind as they conduct their work.

The Network Learning Prototype pulls together these unstructured data mainly text from various sources, including conversations on WhatsApp and Teams, blogs published on UNDP Country Office websites and Medium, and internal reporting feeds. While it is hardly big data, the volume is reaching a point where automation is becoming necessary to detect latent topics and trends that emerge from the network, and to structure and map open-ended taxonomies. The NLP researcher will work on this directly with the Lead Data Scientist of the UNDP Accelerator Labs global team.

The purpose of this procurement exercise is to contract an individual consultant who will work with the lead Data Scientist of the UNDP Accelerators Labs global team on a novel knowledge management pipeline that pulls together unstructured data (mainly text) from various sources including conversations on WhatsApp and Teams, blogs published on UNDP Country Office websites and Medium, and internal reporting feeds and computationally looks for latent topics and trends that emerge from the work being conducted across the network of Accelerator Labs. The main objective of the consultancy will be to improve the topic modeling component of the pipeline.

Duties and Responsibilities

OF WORK, RESPONSIBILITIES AND DESCRIPTION OF THE PROPOSED ANALYTICAL WORK

Explore word and sequence embedding techniques to improve an early prototype for structuring open-ended, multi-language taxonomies.
Design, build, and evaluate topic models and more recent BERT-based zero-shot classifiers in Python for the Network Learning Prototype—primarily in English, but ideally also in French, Spanish, Portuguese, and Arabic.
Collaborate with different teams across UNDP to build a corpus of sustainable development-related documents, presumably in multiple languages, and fine-tune language models using this corpus.
Work with the Lead Data Scientist and the Full Stack Developer to take useful models and classifiers to production and integrate them into a suite of UNDP Accelerator Labs online tools and platforms.
Conduct occasional text analyses for the Accelerator Labs Network global team.
Document the work produced.

Expected outputs and deliverables:

#

Deliverables/ Outputs

Number of working days

Target Due Dates

% of payment

1

Fine tune language models to the corpora of Accelerator Labs documents

50

Apr. 2023

23%

2

Build and maintain a latent feature space for classes (consistent, but unstructured tags used in certain datasets) to map diversity and cohesion of Accelerator Lab activities

50

Apr. 2023

23%

3

Build topic models to feed into the Accelerator Labs Network Learning Strategy

50

July. 2023

22%

4

Explore how zero-shot learning can improve the detection of emergent trends and patterns in the activities of the Accelerator Labs Network

50

Sept. 2023

22%

5

Support ongoing activities related to NLP/ data science, both in the Accelerator Labs Global Team and across the network

10

Nov. 2023

5%

6

Participate in general Accelerator Labs Global Team activities (participate in weekly team meeting, global drop-in calls, etc)

10

Nov. 2023

5%

Competencies

A. Professionalism

Comfortable working with diverse programming languages, and open-source language models, frameworks, and libraries—including Python and PostgreSQL; BERT; SpaCy, scikit-learn, TensorFlow, PyTorch, or fast.ai; Pandas and NumPY; and Matplotlib.
Passionate about natural language processing, data science, and their integration into digital products.
Passionate about open source/ free software movements.
Passionate about the use of data for human and social development.
Ability to work effectively as part of a team, but also independently with little supervision.
Ability and confidence to work directly with partners or clients to define requirements.
Ability to work with people from around the world, with different backgrounds, motivations, and competencies.

B. Planning and organization

Ability to meet deadlines, work under pressure, manage workflows, and operate as part of a distributed team with members across almost every time-zone on the planet.
Ability to prioritize activities and assignments.
Possesses good organizational skills.

C. Knowledge management and learnings

Motivation to continuously learn new things, and ability to put them to use.
Motivation to share personal knowledge and experience with others.

D. Communication

Speaks and writes clearly and effectively, with a particular attention for the audience.
Demonstrates openness in sharing information.
Listens to others.

Required Skills and Experience

Academic qualifications:

Master's degree or higher in Computational Linguistics, Data Science, Computer Science, Information Retrieval, Statistics, Engineering, or any related field with strong computational and text analysis elements is required.

Experience:

A minimum of three years of experience (this can include time as a PhD student) in building and training language models (e.g., recurrent neural networks, transformers, etc.) is required.
A minimum of two years of experience in using transformers and pre-trained language models, such as BERT, BART, or GPT; as well as NLP/ text analytics Python libraries, such as SpaCy, scikit-learn, TensorFlow, PyTorch, or fast.ai is required.
A solid online portfolio or repository of NLP work that demonstrates the candidate’s experience is desirable
A peer-reviewed paper track record (Academic publications) is desirable

In line with UNDP’s gender policy, female candidates are highly encouraged to apply.

Language:

Fluency in written and spoken English is required
Working knowledge of French is desirable
Working knowledge of Spanish, Arabic, or any other UN language is an asset.