Megagon Labs

𝗛𝗼𝘄 𝗰𝗮𝗻 𝗮𝗱𝘃𝗮𝗻𝗰𝗲𝗱 𝗡𝗟𝗣 𝘁𝗲𝗰𝗵𝗻𝗶𝗾𝘂𝗲𝘀 𝘁𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺 𝗷𝗼𝗯 𝗺𝗮𝘁𝗰𝗵𝗶𝗻𝗴 𝗮𝗻𝗱 𝗹𝗮𝗯𝗼𝗿 𝗺𝗮𝗿𝗸𝗲𝘁 𝗶𝗻𝘀𝗶𝗴𝗵𝘁𝘀 𝗶𝗻 𝗛𝗥?👇 Information extraction in HR processes textual data like resumes and job descriptions to derive structured insights. This supports recruitment, skill analysis, and workforce planning. Advancements in NLP and ML make it feasible to address large-scale and complex data processing tasks. 🏗️ 𝗦𝗸𝗶𝗹𝗹 𝗘𝘅𝘁𝗿𝗮𝗰𝘁𝗶𝗼𝗻 Skill extraction involves identifying and categorizing competencies from textual data, which is critical for job-resume matching and training program design tasks. Skills include explicit mentions (e.g., "Python programming") and implicit descriptions (e.g., "proficient in data manipulation"). Key resources like ESCO and O*NET offer structured skill taxonomies that underpin skill extraction models. 🪢 𝗧𝗲𝗰𝗵𝗻𝗶𝗾𝘂𝗲𝘀 & 𝗣𝗿𝗼𝗴𝗿𝗲𝘀𝘀 Modern skill extraction leverages annotated resources and advanced ML methods: 𝗔𝗻𝗻𝗼𝘁𝗮𝘁𝗲𝗱 𝗗𝗮𝘁𝗮𝘀𝗲𝘁𝘀: Resources enriched with skill labels (i.e., SkillSpan) effectively train supervised models. LLMs can be used to synthesize annotated sentences to address rare skill identification challenges (e.g., ESCO Skill Sentences). 𝗡𝗘𝗥 𝗠𝗼𝗱𝗲𝗹𝘀: Transformer-based architectures like BERT excel in identifying context nuanced skills. 𝗚𝗿𝗮𝗽𝗵 𝗡𝗲𝘂𝗿𝗮𝗹 𝗡𝗲𝘁𝘄𝗼𝗿𝗸𝘀: GNNs capture relationships between job postings and skills, improving the understanding of implicit connections. 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗧𝗲𝗰𝗵𝗻𝗶𝗾𝘂𝗲𝘀: LLMs synthesize annotated sentences to address rare skill identification challenges. 🥊 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲𝘀 𝗥𝗲𝗺𝗮𝗶𝗻 𝗔𝗺𝗯𝗶𝗴𝘂𝗶𝘁𝘆: Terms like "engineering" can be vague or ambiguous due to the different types of engineering present across domains, i.e., software, civil, and chemical. 𝗨𝗻𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲𝗱 𝗗𝗮𝘁𝗮: Resumes and job postings lack consistent format and often include irrelevant information. 𝗘𝘃𝗼𝗹𝘃𝗶𝗻𝗴 𝗦𝗸𝗶𝗹𝗹𝘀: Emerging technologies require constant updates to skill taxonomies. 𝗕𝗶𝗮𝘀: Systems may amplify biases present in training data, underrepresenting certain demographics or non-traditional qualifications. 📱 𝗔𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀 𝗮𝗻𝗱 𝗜𝗺𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀 Skill extraction supports job matching, labor market analysis, and personalized training, enabling better alignment between candidates and job roles while revealing macroeconomic trends in skill demand. 🚀 𝗞𝗲𝘆 𝗔𝗿𝗲𝗮𝘀 𝗳𝗼𝗿 𝗙𝘂𝘁𝘂𝗿𝗲 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁 𝗗𝗼𝗺𝗮𝗶𝗻-𝗦𝗽𝗲𝗰𝗶𝗳𝗶𝗰 𝗣𝗿𝗲𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴: Adapting LLMs to HR tasks improves precision. 𝗗𝘆𝗻𝗮𝗺𝗶𝗰 𝗧𝗮𝘅𝗼𝗻𝗼𝗺𝗶𝗲𝘀: Updating models with real-time labor market data. 𝗖𝗿𝗼𝘀𝘀-𝗗𝗼𝗺𝗮𝗶𝗻 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴: Borrowing techniques from other fields. Survey on NLP for HR by Naoki Otani, Nikita Bhutani, & Estevam Hruschka: https://shorturl.at/ffL8F #AI #MachineLearning #NLP #NLP4HR #jobs #LLM #data #ESCO

Join us on our research endeavors!

🤔 𝗛𝗼𝘄 𝗰𝗮𝗻 𝗮𝗱𝘃𝗮𝗻𝗰𝗲𝗱 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝗿𝗲𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻𝗶𝘇𝗲 𝘁𝗵𝗲 𝘄𝗮𝘆 𝘄𝗲 𝗶𝗻𝘁𝗲𝗿𝗽𝗿𝗲𝘁 𝗿𝗲𝘀𝘂𝗺𝗲𝘀, 𝗷𝗼𝗯 𝗽𝗼𝘀𝘁𝗶𝗻𝗴𝘀, 𝗮𝗻𝗱 𝗷𝗼𝗯 𝘁𝗶𝘁𝗹𝗲𝘀 𝗶𝗻 𝗛𝗥? Language understanding is critical in transforming human resources (HR) processes, serving as a foundation for many downstream applications. Some of the key tasks in the HR domain include job title normalization and the interpretation of resumes and job postings—two areas where semantic and discourse-level understanding are paramount. ✨ 𝗝𝗼𝗯 𝗧𝗶𝘁𝗹𝗲 𝗡𝗼𝗿𝗺𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 Job title normalization involves identifying the semantic meaning behind diverse job titles and mapping them to standardized categories. This task is complicated by the variability and ambiguity of job titles across industries and companies. For example, a “Software Engineer” may include varying levels of seniority, functional areas, and additional descriptors like “Remote” or “Backend Specialist.” Techniques such as incorporating behavioral data (e.g., career trajectories) and compositional modeling of job titles enhance this process. These advancements make it easier to align job titles with predefined taxonomies like O*NET or international classification systems. ✨ 𝗥𝗲𝘀𝘂𝗺𝗲 𝗮𝗻𝗱 𝗝𝗼𝗯 𝗣𝗼𝘀𝘁𝗶𝗻𝗴 𝗔𝗻𝗮𝗹𝘆𝘀𝗶𝘀 Understanding resumes and job postings involves segmenting, reorganizing, and classifying free-form and often unstructured data. These documents frequently mix information such as duties, skills, and benefits, necessitating advanced semantic segmentation techniques. Properly extracting relevant sections significantly improves applications like candidate-job matching and skill extraction. Both tasks demonstrate the intersection of HR needs and natural language processing (NLP) techniques. Document classification, phrase composition, and rhetorical role labeling are just a few of the NLP methods enhancing these HR applications. Conversely, challenges in the HR domain—like lexical ambiguity and document structure—spur innovation in NLP research, fostering new approaches for better text understanding. 💫 𝗪𝗵𝘆 𝗜𝘁 𝗠𝗮𝘁𝘁𝗲𝗿𝘀 By addressing these language understanding challenges, HR systems can achieve more accurate and meaningful analyses, leading to better hiring decisions and improved workplace dynamics. It’s a powerful example of how technology transforms traditional practices into smarter, more efficient processes. This is just a section of our NLP for HR survey paper, to read the full paper go to https://lnkd.in/gffRwD6u #NLP #MachineLearning #HR #jobs #LLMs Naoki Otani Estevam Hruschka

We’d like to thank Bhavana Dalvi, Lead Research Scientist at Ai2, for a wonderful presentation on Continual Learning Agents with Dynamic Memory: Case Studies and Methodologies. #womenintech #LLM #Agents #research #AI #AI2 #computerscience #MachineLearning #NLP

📣 We are Accepting Applications for Summer 2025 Internships! We offer research opportunities for PhD students who conduct research on LLMs, NLP, HCI, Machine Learning, Data Science, and more.  Join us this summer to collaborate with industry experts on a research project, and expand your knowledge while contributing to cutting-edge advancements in AI. Apply today! #internships #intern #summerinternship #phdlife #ai #research #phd #machinelearning #LLM #NLP #masters #engineering #EMNLP #ACL

Are you attending #EMNLP2024? Come and learn about our research and our career and internship opportunities. #LLM #AI #phdlife #internships #careers #MachineLearning #PhD

🚀 Want to improve your LLM responses? Read our tutorial for implementing AmbigNLG! Addressing task ambiguity in Natural Language Generation to drive more accurate, context-aligned outputs. https://lnkd.in/gG5nFEij 🤖AmbigNLG makes it easier for LLMs to follow more precise instructions, particularly in tasks like summarization, question generation, and dialogue creation. Ambiguous instructions are a major barrier to getting high-quality results from LLMs, and our method addresses this head-on. We also follow a human-in-the-loop approach so users can interactively clarify instructions and improve outcomes. As many of you know, even the most advanced Large Language Models (LLMs) struggle when given unclear or ambiguous instructions, leading to poor or misaligned results. AmbigNLG changes that. What We’ve Developed: 💠 Ambiguity Taxonomy: We build a taxonomy to categorize the major ambiguities that appear in real-world instructions for LLMs—things like missing context, unclear keywords, or unspecified lengths. 💠 AmbigSNI NLG Dataset: To support this work, we built a dataset of 2,500 annotated examples that help refine NLG instructions and improve the clarity of text generation. 💠 Better Outputs from LLMs: By using our method to refine initial instructions, we’ve seen up to a 15-point improvement in how closely generated text matches user expectations, measured by ROUGE scores. ⚡️Learn to implement AmbigNLG in your LLM, read the tutorial: https://lnkd.in/gG5nFEij Read the research paper, published at #EMNLP2024 and written by Ayana Niwa & Hayate Iso: https://lnkd.in/gFjGGGYC #AmbigNLG #NLP #tutorial #LLMs #AI #MLEngineering #EMNLP #phd

Attending #EMNLP2024? Don't miss Hayate Iso's poster presentation on "AmbigNLG: Addressing Task Ambiguity in Instruction for Natural Language Generation". Poster session 04 at 4pm EST. #research #data #LLM #NLG #NLP #phd

If you attending #emnlp2024, pass by Megagon Labs booth and let’s chat more about this and other exciting #nlproc, #llms, #artificialintelligence projects!!!

🔈 Now Accepting Applications for Summer 2025 Internships! Apply Now: https://lnkd.in/gv6Z7SJk Megagon Labs has exciting research opportunities for PhD students who conduct research on LLMs, NLP, HCI, Machine Learning, Data Science, and more. Join us this summer to gain hands-on experience, collaborate with us on real projects, and learn from industry professionals. Apply today! #LLMs #NLP #HCI #MachineLearning #DataScience #AI #PhD #Internships #Intern #PhDLife #SummerInternship #database