KIBITKIBIT’s hallmarks and future direction

01KIBIT helps you realize what you have not been able to do

Emulate human decision-making to analyze vast amounts of data

KIBIT learns the specific features of your training data, and evaluates other, unseen data based on the result of its learning. This enables KIBIT to extract the target information that meets your decision criteria from the ocean of data.
Specifically, in its learning phase, KIBIT uses morphological analysis to identify parts of speech and words within unstructured, free-form textual data. Next, KIBIT ranks the importance of each word and sentence based on the criteria you set, optimizing the feature axis. In the evaluation phase, KIBIT weighs (attaches scores to) words in unseen data reflecting the relevance of each word based on the optimized feature axis. Finally, it displays the results ranked by how well they match the user’s criteria.

We call our proprietary learning and evaluation algorithm “Landscaping". Landscaping is at the heart of our KIBIT suite of AI technologies. The name derives from the similarities we see between the physical process of creating gardens from the contours of nature (landscaping) and our vision for the ideal process and characteristics of data analysis.
Using this algorithm, KIBIT can automatically learn the subtle nuances of human decision-making. The result of learning can be applied to various business fields. This endows KIBIT with a general utility; for example, it is used to perform predictive coding—a method to identify documents relevant to a specific litigation case by evaluating the relations between the documents and the case.

Document sample in investigation file group. Assesses whether documents are key to the investigation. Criteria used to pinpoint relevant documents from the remaining file group. Replaces human experts by learning from their tacit knowledge (subtleties of human behavior)

Down-to-earth solution: requires only a small amount of training data

In the same way that people struggle to estimate how much they need to study to pass an exam, it is hard to know how much learning an AI-based system requires in order to operate effectively. Why is this? Because the volume of learning required for AI-based systems varies depending on complex factors, including the purpose, quality of the data used, and expected performance. KIBIT is different though. It can achieve high levels of performance with minimal learning, because performance improves through relearning processes even when the data volume is limited.

KIBIT conducts weight refinement based on characteristics of data distribution. Automatically improves the completeness (recall rate) of target information.

02Enrich your decision-making process

Deliver recommendations that fit your tastes

Big data in the digital marketing field includes text-based information on websites. By analyzing this big data, KIBIT can summarize comments made on review websites and provide online users with valuable recommendations about products and services they may not have noticed themselves.
Specifically, KIBIT learns your unexpressed needs from textual information on websites you specify, then applies the results of that learning to countless websites to dig out more information – which may contain gold nuggets matching your unexpressed needs. Just the simple inputs are all it needs to find the information you desire from the Internet—interesting, and sometimes even completely unexpected, information.

User “likes” an arbitrary review of a restaurant that matches their preferences. KIBIT analyzes reviews liked by the user, learning their preferences. KIBIT uses learning to analyze other online reviews, helping to identify restaurants that match user preferences.KIBIT recommends restaurants; learning can also be used to make recommendations in other categories.

Provide reasoning

KIBIT dynamically creates content that is fed back to users. This content includes explanations supporting its data evaluation, as well as visual aids for the distribution rate of topics within multiple data sets. Suppose you add a piano piece of a composer to your Favorites folder. KIBIT may then recommend an opera to you, but it also explains that this was because the composer of the opera is the same.
In order to deepen the interaction between you and KIBIT, we think it is important that KIBIT provides explanations as to why the connection between the suggested piece of information and your preferences was found. We are developing future interfaces to make this scenario a reality.

Conversation with AI

When you start using software specialized in high level data processing, you will be required to learn how to use it and actively act upon it so that you can find and interpret data obtained by the system. Such a system requires a certain level of literacy for configuration and tuning. It is difficult for untrained people to obtain information outside their literacy range.
We are addressing this challenge by equipping KIBIT with an interactive agent. For example, if the AI cannot find what you want, it will provide you with the reasoning for it, and offer you alternatives. Thus, KIBIT can drill down into your needs during the interaction, allowing you to track information beyond your literacy. This will give you a brand new user experience where you feel assured and obtain an in-depth understanding of what KIBIT is searching for.

Our robot will be your partner

For effective operation, an AI requires a platform pertinent to its purpose. Simply displaying the AI avatar on your screen may not result in a smooth conversation, because it is not very easy for humans to recognize it as an object you can interact with. We need to solve this communication hurdle to encourage users to actively collaborate with the AI, making optimal use of its performance.
We have developed a communication robot “Kibiro” equipped with KIBIT which is now on sale. Kibiro has an engaging appearance and a charming and natural response ability for easy interaction. Kibiro can learn your tastes from information acquired through user interaction, and become your personal partner who can even uncover unexpressed facets of your personality.

※Check out our “FRONTEO Research and Development Report”for more details about KIBIT.

Analyzing Big Data and looking into human behavior.Behavior Informatics Laboratories