How to classify documents automatically using NLP
0 likes108 views
This document provides an overview of how to classify documents automatically using natural language processing (NLP). It begins with introducing NLP text classification and the types of classification that can be performed at the document, paragraph, sentence, and sub-sentence levels. It then discusses several business applications of content classification including legal document discovery, enabling customer support, and online content classification. The document demonstrates a live classification of news articles into categories. It also discusses challenges of implementing AI/ML projects and best practices for data collection, quality, security, labeling, infrastructure, skills, speed and continuous improvement. It promotes the capabilities of Skyl.ai as an ML automation platform to help overcome these challenges.
How to classify documents automatically using NLP
- 1. 1 How to classify documents automatically using NLP
- 2. Technology leader with 20+ years expertise in Product Development, Business strategy and Artificial Intelligence acceleration. Active contributor in the New York AI community Extensively worked with global organizations in BFSI, Healthcare, Insurance, Manufacturing, Retail and Ecommerce to define and implement AI strategies Nisha Shoukath Co-founder, People10 & Skyl.ai The Speaker
- 3. Extensive experience building future tech products using Machine Learning and Artificial Intelligence. Areas of expertise includes Deep Learning, Data Analysis, full stack development and building world class products in ecommerce, travel and healthcare sector. Shruti Tanwar Lead - Data Science The Speaker
- 4. Technology enthusiast with 13+ years of experience working in the information technology and services industry. Leads cutting- edge solutions for businesses using Machine Learning and Artificial Intelligence. Areas of expertise includes Architecture design, Solutioning, Data Engineering and Deep Learning.Mohit Juneja Solutions Architect The Speaker
- 5. Bikash Sharma CTO and Co-founder at Skyl.ai CTO & Software Architect with 15 years of experience working at the forefront of cutting-edge technology leading innovative projects Areas of expertise include Architecture design, rapid product development, Deep Learning and Data Analysis The Panelist
- 6. Getting familiar with ‘Zoom’ All dial-in participants will be muted to enable the presenters to speak without interruption Questions can be submitted via Zoom Questions chat window and will be addressed at the end during Q&A The recording will be emailed to you after the webinar Please familiarize yourself with the Zoom ‘Control Panel’ on your screen
- 7. Live demo of News category classification using NLP ...In the next 45 minutes Content classification and how businesses are leveraging it 1 2 3 How to quickly overcome the challenges in building ML models
- 8. Machine Learning automation platform for unstructured data A quick intro about Skyl.ai Guided Machine Learning Workflow Build & deploy ML models faster on unstructured data Collaborative Data Collection & Labelling Easy-to-use & scalable AI SaaS platform
- 9. POLL #1 At what stage of Machine learning adoption your organization is at? ⊚ Exploring - Curious about it ⊚ Planning - Creating AI/ML strategy ⊚ Experimenting - Building proof of concepts ⊚ Scaling up - Some departments are using it ⊚ In production - Using it in product features ⊚ Transforming - AI/Ml driven business
- 10. Content Classification & its applications01
- 11. Text Classification is the activity of labeling natural language texts with relevant categories from a predefined set. Text Classification
- 12. Types of text classification Text analysis can be performed at: ⊚ Document level - obtain relevant information for a full document. ⊚ Paragraph level - extract the most important categories of a paragraph ⊚ Sentence level - get relevant information from single sentence ⊚ Sub-sentence level - obtains relevant information of sub-expressions within sentences
- 13. Document Classification Why do we need it? ⊚ Extracting insights from large volume of unstructured data like articles, survey responses, or support tickets is complex ⊚ Time consuming manual classification method ⊚ Traditional rule based system can’t handle changing data
- 14. Business applications of Content Classification
- 15. Legal Document Discovery Find the relevant documents in the fastest way possible ⊚ File Type Identification ⊚ Sensitive Data Tagging ⊚ Identify the language Source - Case Central
- 16. Laura Amy Sam Jessika Enabling Customer Support Email & query management ⊚ Scan and redirect Emails to the right office/department ⊚ Filter spam ⊚ Identify customer issues with social listening and ticketing
- 17. Online Content Classification Provide better user experience ⊚ Tagging content/news or products using categories as a way to improve browsing ⊚ Identify related content on website Business Sports Entertainment Politics Technology News
- 18. Live Demo of news category classification02
- 19. Approach
- 20. Live Demo of news category classification
- 21. Skyl.ai - as ML automation platform Efficient Data Management Solve your data issues; collect and manage data efficiently Accuracy & Quality Maintain accuracy and quality; train and test faster; monitor quality Effective Collaboration Collaborate and manage projects efficiently Early Visibility Get early visibility; visualize and affirm correctness on every step of the way Scalable High - Performance Access on-demand and scalable, high-performance infrastructure Reduce Cost Reduce cost of implementation; do it with less specialized resources
- 22. POLL #2 Some challenges that you are facing while implementing AI & Machine Learning ⊚ Not started yet, so no challenges ⊚ Data collection ⊚ Data Labeling ⊚ Large volumes of data ⊚ Identifying the right data set to train ⊚ Data Security ⊚ Lack of knowledge of ML tools ⊚ Lack of end to end platform ⊚ Lack of expertise ⊚ Choosing the right algorithms
- 23. Overcoming the AI / ML challenges with the right tools and technologies03
- 24. Best Practices for Data Collection ⊚ Use relevant data sources for data collection ⊚ Establish proper data collection mechanisms ⊚ Do not stop with too-small data sample size Data Collection Data Quality Data SecurityData Security Data SecurityData Labeling
- 25. Best Practices for Data Quality ⊚ Do validate your data and data sources ⊚ Clean up your data regularly - “garbage out” ⊚ Data correction - remove duplicates, missing data, etc ⊚ Check the consistency of data while data acquisition Data Collection Data Quality Data SecurityData Security Data SecurityData Labeling
- 26. Best Practices for Data Security ⊚ Monitor data processes continuously to mitigate risks ⊚ Increase data security with encryption and tokenization ⊚ Controlled access flows with different organizational roles Data Quality Data Collection Data SecurityData Security Data SecurityData Labeling
- 27. Best Practices for Data Labeling ⊚ Define the problem you want to solve and use relevant labels inline with the entities you want to predict ⊚ Analyse trends and progress of your data labeling in real time - to find biases ⊚ Do not add new entity types midway ⊚ Use short tag lists and annotationsData Labeling Data Quality Data SecurityData Security Data Collection
- 28. Challenges ⊚ Requisite Infrastructure ⊚ Cost of Infrastructure ⊚ Data and ML pipeline ⊚ Model at scale for inference Best Practices ⊚ Use SaaS Model (Pay as you go) - reliable, scalable and secure ⊚ The right software tuned and optimized to fit the underlying hardware ⊚ A flexible infrastructure that can be deployed in the cloud or in an on-premise data center to optimize performance Technology issues and solutions
- 29. Best Practices ⊚ Train existing employees with education related to AI and ML ⊚ Use Saas products with good documentation, support and implementation that alleviates the need to have highly skilled data scientists and resources with multiple skills. 40% Lack of skilled talent Source: Techrepublic Barrier in adopting AI ⊚ Companies face shortage of necessary in- house talent. Specialized skills and knowledge
- 30. Challenges ⊚ Long implementation time ⊚ Measure ROI of the AI deployment Best Practices ⊚ AI implementation results in increased process efficiency and automation. ⊚ Create own AI KPIs and analyze the difference in the measurements before and after AI deployment. TechRepublic claims that 56% of global CEOs expect it to take 3-5 years to see any real ROI on their AI investment. Speed and time to market
- 31. Collect Feedback Monitor the model Process Feedback Deploy the changes Train and Evaluate Continuous Improvement Best Practices ⊚ Perform incremental and measurable improvements ⊚ Monitor your deployed models and analyse inference count, accuracy and execution time. ⊚ Check model performance in real time Monitoring and continuous improvement
- 32. AI Project Management More Challenges and Concerns ⊚ Project Cost ⊚ Return on Investment ⊚ On-demand scalability ⊚ Iterative corrections in AI project Source: AI for People and Business: A Framework for Better Human Experiences and Business Success DATA Time Cost Performance Requirements The TCPR Model
- 33. Skyl.ai - as ML automation platform Efficient Data Management Solve your data issues; collect and manage data efficiently Accuracy & Quality Maintain accuracy and quality; train and test faster; monitor quality Effective Collaboration Collaborate and manage projects efficiently Early Visibility Get early visibility; visualize and affirm correctness on every step of the way Scalable High - Performance Access on-demand and scalable, high-performance infrastructure Reduce Cost Reduce cost of implementation; do it with less specialized resources
- 34. ⊚ Free 1 month Trial + POC ⊚ Complimentary 30 min consultation ⊚ AI Implementation Playbook www.skyl.ai [email protected] Special offer for you...
- 35. ? Questions?
- 36. 36 85 Broad Street, New York, NY, 10004 +1 718 300 2104, +1 646 202 9343 [email protected] We hope to hear from you soon Thank you for joining!
Editor's Notes
- #2: Hello everyone and welcome. Thank you for joining today’s webinar on How to classify documents automatically using NLP. My name is Edwin Martinez and I’ll be your host today. First off, I’d like to introduce 3 speakers for today’s webinar, who are experts in the field of AI.
- #3: First we have Nisha Shoukath - Nisha is a technology entrepreneur with background in investment banking. She’s co-founded two successful technology startups and has worked with wide variety of global organizations from different industries. She has an active presence in the AI Community and helps Enterprises with defining AI strategy, and AI roadmaps. Welcome, Nisha!
- #4: Next we have Shruti Tanwar - Shruti is an expert in data science who is a veteran in building SaaS products using Machine Learning and AI. Her expertise includes Deep Learning and Data Analysis, as well as full stack development and building tech products in various different fields such as ecommerce, travel, and healthcare. Welcome, Shruti!
- #5: Next we have Mohit Juneja, Mohit is a Solutions Architect and Technology supporter with over 13 years of experience in the IT and Service industry. He leads cutting-edge solutions for businesses using Machine Learning and AI. He’s an expert in Architect design, Data Engineering, and Deep Learning. Welcome Mohit!
- #6: Finally, we have Bikash Sharma, joining us as a panelist. Bikash is CTO and Software Architect with 15 years of experience in leading innovative software projects and solutions. He’s co-founded Skyl with his expert knowledge in AI and Machine Learning. Welcome, Bikash!
- #7: Before we begin, I’d like to briefly talk about some Zoom features that will be relevant to us. All participants in the webinar will be muted to avoid any interruptions during the session. Any questions you might have can be submitted to the Zoom Questions chat window in the control panel, located on the bottom of the screen. We’ll make sure to address your questions during the Q&A session. Also, the recording of the webinar will be emailed to you afterwards, just in case you’ve missed any talking points or wish to view it again. So that’s all for the introduction - let’s bring in our first speaker Nisha for more on this topic
- #10: Exploring - Curious about it Planning - Creating AI/ML strategy Experimenting - Building proof of concepts Scaling up - Some departments are using it In production - Using it in product features Transforming - AI/Ml driven business
- #12: One of the first studies on Twitter data for sentiment was to study public perception of Obama’s performance as President. Another example could be the to explore the variation of sentiment regarding the TV series “Game of Thrones.” The unpredictable episode “The Rains of Castamere” resulted in a lot of negative tweets and a peak in the sentiment score.
- #13: So, which one is better? Should you analyze your documents as a whole or break them into smaller units? Unfortunately, there is no straight answer. Your choice will depend on your data and objectives.
- #14: In the past, many companies have used traditional business intelligence tools to monitor social media. However, this is not efficient because traditional BI tools cannot handle true sentiment analysis, capture sarcasm, or process and learn new slang.
- #17: For example, let’s say you work for a software company and that you use document classification to tag incoming support tickets. You can define that new tickets labeled as Bug should be automatically routed to the technical team.
- #19: How
- #23: Not started yet, so no challenges Data collection Data Labeling Data Bias Large volumes of data Identifying the right data set to train Lack of knowledge of ML tools Lack of end to end platform Lack of expertise Choosing the right algorithms Monitoring the model performance
- #24: Benefit
- #25: Data is one of the most valuable resources today’s businesses have. The more information you have about your customers, the better you can understand their interests, wants and needs. Use of relevant data sources - to get consistent and accurate data relevant for the problem you want to solve Collection mechanisms - A formal data collection process is necessary as it ensures that the data gathered are both defined and accurate. Small sample size - do not give the distribution of the data to the edge cases and will not train the model for exceptional cases. Keep in mind that machine learning is a process of induction. The model can only capture what it has seen. If your training data does not include edge cases, they will very likely not be supported by the model.
- #26: The best practices can be achieved by Data Cleaning: Applying a detailed data analysis at the initial phase for recognizing which sorts of irregularities and errors must be expelled. Notwithstanding a manual assessment of the information or data samples, analytic programs are frequently expected to pick up metadata about the data resources and distinguish the issues of data quality. Don’t let bad data or records go unresolved - remove duplicates and fill missing data; For missing data, you should flag and fill the values. Flag the observation with an indicator variable of missingness. Incorrect or inconsistent data leads to false conclusions. And so, how well you clean and understand the data has a high impact on the quality of the results.
- #27: Encrypted data sources All data sources are encrypted; thus giving users an additional layer of security, making sure your data stays safe and protected. Access controlled flow Defined and controlled access flows with different organizational roles like business owner, project lead, collaborators etc. allow for selective restriction so that you have full command to regulate who can view or use resources in your ML projects.
- #28: Adding tags midway - For example, the set of tags for a pizza chatbot might start with the tags “Size” “topping” and “drink” before someone realizes that you also need a “Side Dish” tag to capture Garlic Bread and Chicken Wings. Simply adding these tags and continuing work on the documents that haven’t been labeled yet poses a danger to the project. The new tags will be missing from all of the documents annotated before the new tags were added This means that your test set will be wrong for those tags, and your training data won’t contain the new tags leading to a model that won’t capture them. <https://towardsdatascience.com/four-mistakes-you-make-when-labeling-data-7e431c4438a2> In an annotation process, increasing the number of choices the annotator needs to make slows them down and leads to poor data quality.
- #29: Requisite Infrastructure: When launching a machine learning initiative, organizations can easily underestimate the resources they need for infrastructure. There can be substantial infrastructure requirements for machine learning, especially in the cases of image, video, and audio processing. Cost of Infrastructure: Training and deploying a scalable infrastructure to support machine learning can be expensive and difficult to maintain. Having a cloud approach allows experimentation with machine learning at scale without the overhead of physical hardware acquisition, configuration, and deployment. Data & ML pipeline: AI, Machine learning and deep learning solutions require a high degree of computation speeds offered. Model at Scale for inference: Deploying a scalable infrastructure to support machine learning can be expensive and difficult to maintain. Things get tedious and difficult to maintain at scale compared to having single server deployment or in a developer’s environment which is not a usual case.
- #30: Skills challenges: Choice of right ML algorithm - ML, DL AI product management - Dealing with Cold start, managing data labeling project, keeping transparency in the project; Keeping the model up to date. Adoption of AI technologies requires specialists like data scientists, data engineers, infrastructure engineers and other SMEs (Subject Matter Experts).
- #31: Even with long implementation time, AI has potential to cut the expenses. TechRepublic claims that 56% of global CEOs expect it to take 3-5 years to see any real ROI on their AI investment. Machine automation produces quality products faster and more efficiently, while providing critical information to help managers make more informed business decisions.
- #32: Making continuous improvement part of company culture is an excellent and cost-effective approach to tackling an organization’s most difficult challenges. When supported by improvement technology, results can be achieved quickly and success can be sustained over time.
- #33: On-demand scalabilty: The truth that it’s better to have a working prototype of a smaller product, rather than an unfinished large one, still stands here with machine learning products. New ML MVPs should be prioritized based on the speed of delivery and their value to the company. If you can deliver products, even those which may be smaller, with speed, it can be a good, quick win for the whole team—you should prioritize these products first. Organizations need to keep in mind that machine learning is an iterative process, and modifications to models might happen over time to support changing requirements. TCPR Model: The TCPR model represents an indeterminate system—one in which more than one solution exists. Notice that the TCPR model rests on a foundation of data. This is critical. There’s no point even talking about the four components of TCPR without first identifying what data sources and fields (aka attributes or features) are available. Link: TCPR
- #35: Thank you Nisha, Mohit and Shruti, for the wonderful presentation and demo. As mentioned earlier, the recording of the webinar will be emailed to you afterwards. [pause] Before we get to the Q&A, I want to mention some of the offers Skyl has for those of you that are curious about incorporating Machine Learning to your business. Skyl offers a free 1 month trial, plus Proof of Concept. You’ll be able to interact with real data on the screen, just like we showed in the demo. You’ll experience the process of going from collecting & labeling the data… all the way to deploying a model! Skyl also offers a complimentary 30 min consultation and an AI Implementation Playbook to go along. This is a great opportunity to see how Skyl can provide Machine Learning solutions to your challenges. If you’re interested in finding out more, please visit the skyl.ai website or you can send an email directly to [email protected].
- #36: Alright, now it’s Q&A time! As a reminder, if you have any questions, go to the question box in your control panel - located on the bottom of your Zoom screen. We’ll try to answer as many questions as possible in the time that we have left. So let’s answer some questions. Sample questions: Shruti - (James) How do I know if my model performance is going down, and how do I fix it? - (Anonymous) How can I know the fairness of a model in Skyl? -(Julie) If I build a lot of models, how do I handle model deployment in that case? -How do you avoid creating a biased model and if you detect one, how do you rebuild it? Nisha -(anonymous) How can Skyl help me with my data labelling needs if I have data privacy issues? Ok, that’s all the time we have for questions today, but feel free to contact us with your specific questions and we’ll make sure to get them answered.
- #37: All right, so we have reached the end of the webinar. We hope you enjoyed it. We have a lot more webinars coming up on different machine learning topics and how they can be implemented into different businesses and industries, So don’t miss out and make sure you sign up for upcoming webinars as well Thank you for joining and I hope you have a wonderful day.