Extractive approaches in automated summarisation
‘Overview of automatic summarisation technology, algorithms and examples of implementation’ describes AI-based summarisation technology.
Automatic summarisation technology is used to summarise large text documents and texts in a short and to the point form, and is widely used in information retrieval, information processing, natural language processing, machine learning and other fields to compress information and facilitate understanding of summarised information. It can be broadly divided into two types: extractive and abstractive summarisation.
In the aforementioned blog, the discussion focuses on extractive summarisation, which is relatively compatible with natural language processing. Abstractive summarisation is a type of summarisation that is based on the concepts of ‘abstract and concrete from the perspective of natural language’ and ‘What is meaning?’. (1) A philosophical approach to meaning and symbols’, it requires an approach that goes into the meaning of the text, making it difficult to apply the mathematical methods that form the basis of machine learning.
If it is just a simple matter of conveying information, as described in ‘Shannon’s Information Theory in a nutshell and reference books’, we are talking about the entropy of information quantity between the conveyer and the receiver. On the other hand, if the conveyor of information is an intelligent organism, as described in ‘The Turing Test, Searle’s Refutation and Artificial Intelligence’, as also described in ‘Life as Information – Purpose and Meaning’, meaning exists behind the behaviour of the organism and is generated by the purpose for which the organism is living. This meaning is generated by the ‘computer’. This meaning exists in a different form from the symbols used in the dialogue, as also described in ‘Dealing with the meaning of symbols in the computer’.
Abstract summarisation involves deducing the semantic information behind this from the symbols (language) being spoken and removing redundant parts from this information.
Here, I would like to consider a qualitative approach to abstraction-type summarisation, based on the ‘one-word summarisation technique’.
The art of summarising in a few words
describes the art of conveying words to others in a shortened and summarised form for ‘those who have something to say but cannot put it into words’.
In the book, it is stated that today’s society can generate and consume in a day the equivalent of a year’s worth of information in the Edo period or a lifetime in the Heian period, and that it is increasingly valuable to ‘think about what is necessary and communicate it’ without being drowned in such a large amount of information.
It lists the following eight processes that can be summed up in a single word
- Have courage (the will to say it): have a strong desire to tell others.
- Know yourself: look at yourself objectively.
- Know the person you want to tell: gather as much information as possible about the person you want to tell.
- Define your destination: Define the destination you want them to reach.
- Find the core: think about what is most appealing to the audience, not just the synopsis.
- Cut down to the bare essentials: too much information is worse than no information at all, so instead of TELLing, focus on the essentials so that the person GETS something.
- Observe how the other person moves: find out if they have GET something properly.
- Developing relationships: the act of communicating is about developing relationships.
In the above algorithm for ‘summarising techniques’, the purpose of communicating is defined as ‘8. the act of communicating is to develop the relationship’, and ‘1. having a strong desire to communicate to the other person’ is maintained as the state for doing so, and then the information to be communicated corresponding to the purpose is defined as ‘5. not a synopsis, but the most attractive part for the other person’. The most attractive part for the other person’, and extracting this information through the processes of “2. knowing yourself”, “3. knowing the other person” and “4. clarifying the destination”, and finally “7. determining whether the other person was able to GET something properly” as feedback on the action.
Communication support using AI technology
In this process, with the exception of steps 1 and 8, AI technologies can be used to support effective self-analysis and communication. The following sections describe generic AI technologies that can be used for each of these steps.
・ Knowing oneself: looking at oneself objectively
– Self-analysis tools: using sentence generation AI (e.g. ChatGPT) utilising NLP (natural language processing) and AI tools based on psychology, it is possible to provide objective feedback on one’s thoughts, opinions and behaviour patterns, and by analysing past data and behaviour history, one’s tendencies, strengths and weaknesses can be Visualisation.
Examples of these could include.
1. a self-analysis system using a sentence-generating AI: Using ChatGPT or other sentence-generating AIs, one can make sentences about one’s past actions, opinions and feelings and obtain feedback on them. For example, if you enter your thoughts into the AI in the form of a diary, the AI will give you feedback on your patterns based on the content, helping you to better understand yourself.
Examples: if you input ‘I have been struggling with my relationships at work recently’, the AI will say, ‘Your problems may be caused by a lack of communication with others. It provides feedback such as ‘Let’s look back at how you are approaching your problems’.
2. behavioural history analysis tools: the AI analyses data from your past behaviour history and decisions to identify trends and patterns (e.g. by inputting task management tools, work records and daily behaviour data into the AI, it can analyse when you work efficiently or what you stress about (e.g. analysing when you work efficiently or what you feel stressed about).
Examples: the AI analyses calendar and task records, provides insights such as ‘you tend to be less focused in the afternoon and are more productive in the morning’ and suggests remedies.
3. psychology-based personalised AI: based on psychological theories, AI provides personality assessment and stress management (e.g. by answering questions based on the Big Five theory and the Myers-Briggs Type Indicator (MBTI), AI provides feedback on your personality traits and behaviour patterns by It provides feedback on your personality traits and behaviour patterns by answering questions based on the Big Five theory, MBTI (Myers-Briggs Type Indicator), etc., and gives advice on which situations cause stress and how to deal with them.
Example: AI diagnoses the user’s personality type through questions and provides feedback such as ‘You are an outgoing and sociable type, but you tend to take time to make decisions’.
4. emotional analysis tools: AI analyses text and voice data to visualise the user’s emotional state and identify changes in stress and motivation that they are unconsciously feeling.
Example: by analysing the content of daily conversations and statements, the AI can say, ‘You seem to be under a lot of stress these days. We suggest you take some time to relax’.
・ Know who you want to communicate with: gather as much information as possible about the person you want to communicate with.
– Personalised AI: Consider AI that can collect and analyse a person’s profile (past interactions and topics of interest) in real-time and provide appropriate messages to them. For example, it could identify a person’s needs and interests through analysis of social media data and feedback.
Examples of these could include.
1. using personalised AI in customer support: by using AI to analyse past interactions with customers, purchase history and support history, it is possible to identify what problems customers are struggling with and provide appropriate support messages. For example, if a customer has repeatedly requested support for a particular product in the past, AI can provide a solution related to that problem in advance.
Example: ‘We’ve re-visited an issue you had with your previous smartphone purchases and we’ve found a solution to your problem. This troubleshooting guide may help.’ Customised support messages can be provided, e.g.: ‘We have reviewed your previous smartphone purchases and found this troubleshooting guide to be useful.
2. personalised AI in marketing campaigns: social media data and customer online behaviour are analysed by AI to automatically generate marketing messages tailored to individual users. For example, if a user has recently posted about a particular product or is interested in a particular brand, relevant offers and content can be provided.
Example: ‘Check out our recent review on the wireless earphones you searched for. Also, special discounts are only valid this week.’ and other promotions based on individual interests are sent.
3. personalised AI in education: by analysing students’ past learning data and performance, AI can provide the best learning content and assignments for each student in real time. This enables support tailored to individual progress and level of understanding.
Example: an AI analyses a student’s test results and says: ‘You seem to be having a little difficulty with problems on the number line. We can provide feedback and remediation, such as ‘You seem to be struggling a little with number sequences, we can provide you with additional practice questions on this topic’.
4. personalised AI to identify the needs of a business partner: AI analyses past communication data (e.g. emails, meeting records, contractual details) with customers and business partners to understand their needs and interests. This enables it to provide them with useful suggestions and information at the right time.
Example: ‘We have prepared detailed material on the new marketing strategy you expressed interest in at our last meeting. We would be happy to review it with you before our next meeting.’ A customised proposal message would be sent to them, such as.
5. influencer targeting on social media: the AI analyses users’ posts and follower behaviour on social media to determine what topics and products they are interested in, and sends personalised ads and messages accordingly.
Example: ‘An influencer you recently followed has also reviewed these new trainers. Try them out and get a special discount.’ Send a message saying.
・ Clarify the destination: clarify the destination you want the person to reach.
– Goal-setting support AI: using machine learning to generate specific action plans using tools that support goal-setting based on past successes and failures, and how to guide the person step by step.
Examples of these could include.
1. professional career planning tools: use machine learning to analyse data on the user’s past work history, skill sets and successful career paths and, based on this, suggest what jobs and career stages the user could move into.
Example: if the user enters ‘I want to become a project manager’, the AI analyses the career paths of successful project managers in the past and suggests ‘as a next step, we recommend attending leadership training and then obtaining the relevant qualifications’, with a specific Provide an action plan.
2. fitness and health management apps: apps that analyse users’ past training data and health status and help them set goals, setting individualised fitness goals based on successful training and eating plans and helping them achieve them step by step.
Example: if a user enters ‘I want to lose 5 kg’, the AI will provide specific steps such as ‘First, I want to lose 1 kg in two weeks, and I recommend a combination of exercise and calorie restriction three times a week’.
3. learning plan generator: analyses the student’s past performance and learning style to generate a learning plan tailored to their goals. It indicates which subjects and topics to focus on based on historical data and has the ability to track progress.
Example: if a user enters ‘I want to improve my maths grade’, the AI will suggest ‘I recommend solving linear algebra problems for 30 minutes every day before the next test. It also provides a specific learning action plan, such as ‘Set aside time at weekends to solve past papers’.
4. business goal-setting tool: analyses historical performance data from companies and teams to help improve operations and set goals. Provide goals and action plans for the next quarter, referring to successful strategies and approaches.
Example: if a team sets out to ‘increase sales by 10%’, the AI will generate an action plan such as ‘Based on historical data, we recommend that you adopt the techniques from marketing campaigns that have been successful in gaining new customers and measure the effectiveness of your campaigns on a monthly basis’.
5. time management support tool: analyses the user’s past time management data to provide an efficient timetable for achieving goals. It suggests which tasks and how much time should be allocated to which tasks and generates daily action plans.
Example: if a user enters ‘I want to complete a new project by the end of the month’, the AI provides a specific timetable, recommending that the user should focus on the project for two hours every day and adjust their schedule to work on the summary at the weekend.
∙ Find the core: think about what is most appealing to them.
– Content optimisation AI: uses deep learning to find the most effective messages and elements for the recipient. Based on past successes and their interest data, it analyses which information resonates with them and presents it to them.
Examples of these could include.
1. optimising marketing content: the AI analyses past marketing campaign data to identify which types of content (e.g. blog posts, videos, social media posts) are most effective for a particular target audience. This allows it to optimise the type and style of content used in the next campaign.
Example: the AI analyses data from a previous campaign for a ‘fashion brand for young people’, and identifies ‘colourful visuals and collaborations with influencers worked well’. We recommend incorporating these elements in your next campaign’.
2. personalising email marketing: analysing users’ past behavioural data (open rates, click rates, purchase history, etc.) to generate the most relevant content and offers for each recipient. This can improve email engagement rates.
Example: an AI can suggest an upsell of a similar product to a ‘customer who has previously purchased a particular product’, such as ‘A new version of your previously purchased product is now available. Generate personalised messages, such as ‘You can buy this product at a special discount’.
3. optimising social media content: analysing engagement data in social media posts to identify which types of posts (e.g. images, videos, texts, Stories) are most effective for specific audiences. This can then optimise the next posting strategy.
Example: if the AI analyses the response of a brand’s followers and decides that video content is getting the highest engagement, it will suggest that ‘next time you should post a short tutorial video’.
4. optimising website content: analysing user behaviour data on the website (e.g. page views, time spent, direct return rate) to identify which elements (e.g. titles, images, CTAs) are resonating with visitors. This can improve the UX/UI of the site and increase conversion rates.
Example: the AI analyses ‘user abandonment on certain product pages’ and provides feedback on ‘recommendations to reduce abandonment rates by making page titles more specific and CTA buttons more prominent’.
5. idea generation for content creation: the AI analyses trend data and search volumes to identify which topics are currently popular. Based on this, it provides writers with ideas for relevant content.
Example: if the AI analyses recent trends and decides that content about eco-friendly products is popular, it suggests writing a blog post with eco-friendly lifestyle tips.
∙ Bussing: narrowing down to the information you need.
– Summarising AI: uses natural language processing technology to summarise information and extract only the necessary parts. For example, complex content can be converted into short summaries to avoid information overload and make it easier for important messages to be conveyed.
Examples of these could include.
1. news summary apps: automatically summarise important information related to the topic of interest to the user from a vast number of news articles. This enables users to grasp the latest news in a short time.
Example: a news summary AI ‘analyses an article on an environmental issue, extracts the main points and provides the user with a summary titled “This week’s environmental news: key policy proposals and their impact”’.
2. business report summarisation tool: summarises the content of business meetings and reports to extract key decisions and next steps. This saves time and gives stakeholders quick access to the information they need.
Example: a summary AI analyses the minutes of a meeting and generates a summary such as ‘Key decision: deadline for Project A extended by two weeks, next progress report 15 Oct’.
3. research article summarisation tool: summarises academic articles and research material, presenting key findings and conclusions in a concise manner. Researchers and students can quickly grasp the vast amount of literature and efficiently obtain the information they need.
Example: an article summarisation AI ‘analyses an article on a biological study and generates a summary that says: “This study shows that a new antiviral drug is effective and further research is needed”’.
4. email summariser: summarises long emails and extracts key messages and action items. The recipient can quickly understand the content of the email and respond quickly.
Example: the summarisation AI ‘analyses the incoming email and generates a summary such as “Next meeting is on 10 October, agenda includes sales report and new product introduction”’.
5. online course summarisers: summarise course content and lectures on online learning platforms to organise key points. Students can review the key points for easier revision.
Example: a summary AI analyses the class recording and provides a summary such as: ‘The main points of this lecture: 1) Basic concepts of economics, 2) Market mechanisms, 3) The relationship between supply and demand’.
6. social media content summarisation: summarises long-form posts and comments on social media and extracts key opinions and sentiments. This enables users to briefly understand the thoughts and opinions of others.
Example: a summary AI analyses a social media post and generates a summary that says: ‘This post highlights the good qualities of the product and states that customer satisfaction is high’.
– Observe how they behave: determine whether they have successfully GET something.
– Feedback analysis AI: analysing the behaviour patterns and reactions of customers and counterparts to check whether the message was effectively conveyed, using AI tools to analyse their reactions in real time, based on sentiment analysis and behaviour prediction.
Examples of these could include.
1. customer support chatbots: analysing customer reactions and satisfaction in real time as they interact with the chatbot. It uses sentiment analysis to assess how satisfied the customer is and whether the problem has been resolved, and has the ability to escalate to a human operator if necessary.
Example: if a chatbot receives a message from a customer stating that the issue has not yet been resolved, the AI analyses the content and determines the need for escalation.
2. product review analysis tool: analyses reviews and ratings posted by customers for a product and performs sentiment analysis. Categorises positive, negative and neutral feedback and identifies which elements are important to the customer.
Example: AI analyses reviews such as ‘the quality of this product is great, but delivery is slow’ and reports positive feedback on improving product quality and negative feedback on delivery.
3. analysing the effectiveness of online advertising: tracking the behaviour of the audience of an advertisement and analysing their response to the advertisement. It analyses click rates, conversion rates and user sentiment to identify which elements are most effective.
Example: the AI analyses data from the ad campaign and provides feedback such as ‘this ad banner has the highest click rate and most positive comments’.
4. social media sentiment analysis: analyses social media posts and comments relevant to the brand to assess sentiment towards the brand. Positive and negative responses are tracked in real time and strategies adjusted as necessary.
Example: the AI analyses posts against the brand and reports that ‘feedback on recent product releases has been positive and highly supportive of the brand from customers’.
5. employee feedback analysis: feedback and surveys provided anonymously by employees are analysed to assess the workplace atmosphere and employee satisfaction. Based on sentiment analysis, the AI can understand how employees feel and suggest improvements.
Example: the AI analyses employee feedback and provides insight into ‘there is a lot of dissatisfaction with the work environment and improvements are needed’.
6. e-learning platform feedback analytics: analysing learner behaviour and evaluations to identify which materials and topics were most effective. Real-time tracking of learner responses to help improve educational programmes.
Example: the AI tracks learner progress and provides feedback that ‘this topic is too challenging for many students and needs additional support’.
These AI technologies will improve the quality of communication and support the creation of a state of ‘getting across’ as well as ‘communicating’ to others.
reference book
Reference books relevant to the abstract approach and AI assisted communication in summaries include.
1. ‘Artificial Intelligence: A Guide to Intelligent Systems’ by Michael Negnevitsky.
– Contains commentary on the fundamentals and applications of AI systems and also covers technologies for communication support.
2. ‘’.
– Explains the basic concepts and techniques of natural language processing (NLP) and provides insight into how AI can support communication.
3.‘The Art of Thinking Clearly’ by Rolf Dobelli
– Introduces a way of thinking to understand complex information in a concise way, providing useful perspectives on summarising and abstracting.
4. ‘The Design of Everyday Things’ by Don Norman
– Presents theory and practice on user interface and communication design, providing useful information for improving the user experience of AI systems.
5. ‘A Beginner’s Guide to Natural Language Processing’
– Introduces the basics of natural language processing technology and provides practical examples.
6. ‘Information Theory, Inference, and Learning Algorithms’ by David J.C. MacKay
– Deals with information theory-based summarisation and abstraction techniques, which can be applied to AI data analysis and communication support.
コメント