What can the modular AI service "KEY RING" do? The possibilities expanding through AI-powered "Natural Language Processing"

Data Artist Inc. focuses on providing digital solutions and solving marketing challenges, primarily through AI. The company developed "KEY RING," a service that builds solutions by combining AI modules. Its key features include ease of use even for non-AI experts and diverse application possibilities through module combinations. Transformation SHOWCASE is publishing a multi-part series on what can be achieved using "KEY RING."

Part 1 explained the development background, overview, and features of "KEY RING." Part 2 introduced case studies primarily utilizing "image processing modules." In this third installment, Satoru Yamamoto of Data Artist introduces "What Natural Language Processing Capabilities Bring." We hope this article inspires you to explore the potential for AI implementation within your own organization.

"KEY RING" is equipped with seven "natural language processing modules"

While AI is now used across various fields, many people likely feel, "I'm interested, but I don't know how to use it" or "It seems difficult to apply it to my company's business." As explained in the first installment featuring our company's Ken Yamada, "KEY RING" provides versatile "generic modules" suitable for diverse projects to address these challenges. This time, we focus on the "natural language processing modules."

The Natural Language Processing Module consists of the following seven components.

・FAQ Module (Chatbot)

・Slogan Generation Module

・Natural Dialogue Module

・Context Understanding Module

・Keyword Networking Module

・Relationship Visualization Module

・Tweet Analysis Module

When it comes to the most compelling feature within our natural language processing system, it would have to be the one leveraging the "Natural Dialogue Module" – the so-called "dialogue engine." This enables an experience where, even though the other party isn't human, it feels like you're having a natural conversation with a person.

We also offer a "Tweet Analysis Module," which we utilize to provide "Twitter Analysis." While this might sound like the "social listening" functionality offered by research or PR firms, our approach is different. By analyzing Tweets with AI, we can estimate key attributes like "whether the poster is male or female, and their approximate age," as well as "whether their sentiment is positive or negative."Since this estimation is derived from vast amounts of Tweet content, it could be described as "visualizing public opinion." Manually analyzing text data has its limitations, so leveraging AI holds significant value.

Natural Language Processing is a fundamental capability for data-driven marketing.

One example of deploying a dialogue engine using a "natural conversation module" is the character " AI Tamiko." Originally developed in conjunction with a TV station's program concept, "AI Tamiko" was created with the ambitious goal of "becoming a beloved AI."Ultimately, she grew to the point of hosting remote drinking parties with fans, achieving considerable success. Furthermore, while being AI, she can have her "character traits adjusted as needed." This allows for collaboration with various companies, enabling her to serve in diverse scenarios such as reception desks, customer service roles, or as an online explainer. Such broad applications are currently under consideration.

Within Dentsu Inc. Japan Network, our AI technology is utilized in a system for predicting TV ratings. It analyzes diverse "program-related data" – such as which talent will appear, recent topics surrounding that talent, trending search keywords, past ratings, broadcast time and genre, and program schedule details – to forecast viewership.

In essence, "natural language processing" can be considered fundamental to data-driven marketing. What is happening in the world right now? What is trending, gaining support, and what is the prevailing atmosphere? Analyzing vast amounts of text data reveals these insights. "Language processing" involves analyzing human speech or text to generate "the most likely next thing to follow."

For example, dialogue engines and chatbots predict and respond with phrases like "this phrase is likely to follow that conversation" based on input language. Their high accuracy makes interactions feel like natural conversation. After extensive learning about word connections, they deduce "under these conditions, this is the most probable outcome."In the context of ratings prediction, it's not simply speculating that "viewership will likely rise because a popular idol is appearing." Instead, it predicts the impact on ratings by considering various factors: "recent discussions about that idol, mentions on social media, media coverage, and news stories." This is essentially marketing itself.

Furthermore, we've recently begun initiatives to directly leverage these capabilities for corporate marketing. Our AI personalization tool, "Microscope," automatically clusters visitors to a company's website based on their past site visit history (using data similarity grouping). It then presents the "most suitable" advertising message to each cluster.Leveraging natural language processing capabilities allows analysis of the content of previously visited sites. This enables the AI to analyze "what type of customer visited our site." Consequently, the advertising messages delivered can also be optimized.

The stage of "building the brain" is over. Now we're moving into the stage of "how to use the brain."

In the past, AI was something you "trained." It required feeding it massive amounts of data for learning before it could be effectively used for prediction. Furthermore, with insufficient data, accuracy suffered, often leading to disappointing results. However, today, the American non-profit AI research organization "OpenAI" has developed AI pre-trained on vast datasets.Data artists have also developed an AI called "GPT-3," which has completed training at a computational cost exceeding 100 million yen, and are beginning to utilize it. We no longer need to train AI from scratch; we have entered the stage of "how to utilize" this pre-trained AI.

For example, using GPT-3, you can simply input a product's specifications, and the AI will write its description. Since it has already learned "what characteristics products with these specifications typically have," it can automatically generate the "most plausible" description inferred from that knowledge.

Applying this further, the AI can also answer questions like "What concerns do people considering a trip to Japan typically have?" To the questioner, it feels like the AI is creating something out of nothing by answering something they didn't know. In reality, the AI analyzes vast amounts of existing data to understand what is commonly associated with keywords like "Japan travel" and "concerns," then returns the most plausible information based on that.In other words, this answer is derived from statistics. For humans, who struggle to process vast amounts of data at once, AI truly provides "predictions beyond our capabilities."

One module in "KEY RING" is the "Slogan Generator." It automatically writes catchphrases for products or services. An example showcasing "the unique charm and shortcomings of AI" emerged here. When I input, "What slogan would make someone want to apply for a card loan?", the response was: "This card loan doesn't require repayment." Well, if you didn't have to pay it back, anyone would want to borrow, right?(laugh). But since a card loan you don't have to repay is impossible, some might say the AI doesn't understand that. Yet, there's something about this that you can't dismiss. If such a "loan you don't have to repay" truly existed, it would surely be the ultimate loan. Once you accept that premise, ideas like "X people won't have to repay in a lottery""No repayment required for the first year," "First-time users don't have to repay" – all sorts of stimulating ideas and strategies become visible from there, and some might actually become viable. However, humans tend to dismiss the idea of a "loan you don't have to repay" as impossible, so they might not even consider that direction from the start. Therefore, while the AI's answer isn't perfect, the direction it points to might hold some possibility close to the correct answer.

The evolution of AI will support Japan's future growth

I hope this discussion has given you some sense of the worldview behind what we call natural language processing. While this strays a bit from "KEY RING," I'd like to conclude by sharing how we plan to evolve this "natural language processing" going forward.

While learning and analyzing text data is advancing, efforts are also progressing to link this text data with image data. If this advances, it will become possible to provide detailed verbal explanations of what an image depicts, and conversely, to automatically generate images based on textual instructions.Efforts in "AI-generated art" are already advancing, making it possible to automatically generate paintings in the style of specific artists. Moving beyond this, AI will eventually be able to draw pictures according to instructions and describe the content of images. In this way, the capabilities of AI continue to expand. I hope you'll also find the evolution of AI interesting.

It's often said that the Industrial Revolution in Britain during the late 18th century created the distinction between "developed nations" and "developing nations," and this trend continues today. The AI evolution we face now may profoundly influence a nation's future growth potential itself. Whether Japan can continue to grow may well depend on its ability to embrace AI's evolution. We believe this and intend to continue engaging with AI's evolution while striving to be of service to you.

This article introduced the "Natural Language Processing Module" within "KEY RING." While the functions and features of each individual module are important, recognizing the fundamental truth that "natural language processing" will become one of the foundational pillars of future marketing could significantly shift your perspective and expectations regarding AI.

Transformation SHOWCASE will continue to showcase AI solution cases built using this "KEY RING." We encourage you to read those articles as well.