KIBIT, AI Engine of FRONTEO
01Landscaping, the core technology of KIBIT
Of the AI-related technologies incorporated in KIBIT, FRONTEO refers to the core technology as “landscaping.” This is because the technology, which deftly identifies the specific features of a small volume of data and makes judgement that can be consistently applicable to a larger volume of unknown data, functions in a manner akin to landscaping, or visualizing a complete scene from a narrow field of view.
Here is a brief explanation of the “landscaping” technology.
When analyzing text data, KIBIT identifies the parses (word classes) and extracts certain words to calculate the degree of importance for each extracted word in relation to the targeted information. In calculating the degree of importance, KIBIT uses the concept of transinformation to determine whether the word is relevant.
Next, KIBIT compiles the degree of importance for each word within the text to assign a score, realigning data by score in a descending order. These processes are performed in “landscaping,” an algorithm independently developed by FRONTEO as a method of learning and inference for KIBIT. This algorithm allows KIBIT to learn the tacit knowledge of experts – or knowledge based on the experiences and judgment of humans.
02Learning “tacit knowledge” – a special feature of KIBIT
KIBIT can learn from a small volume of text data what is called “tacit knowledge,” or knowledge based on the experiences and judgement of humans. This ability makes it possible to streamline such operations as checking documents of massive volume, which requires judgement of humans, to significantly reduce operational burdens.
Let’s take financial institutions, for example. When these entities sell financial instruments, various laws and regulations must be abided by, including the Financial Instruments and Exchange Act, voluntary regulations of the Japan Securities Dealers Association and the Insurance Business Act. Financial institutions also must check several items, such as the appropriateness of transactions in proportion to customers’ willingness to purchase, knowledge, investment experience, allocation ratios and other factors, inadequacies in the documents offered to customers, consent by customers’ family members, and any unreasonable aspects found in customers in their discussions.
However, human eyes cannot check all cases, and judgment criteria may differ among persons in charge according to their skills and experiences. Moreover, the quality of checks may not be secured because of possible human errors due to fatigue, among other reasons.
Here is a case in which KIBIT is used for a financial institution.
The Financial Instruments and Exchange Act prohibits financial institutions from conducting inappropriate solicitation and calls for the necessity of protecting investors. Financial institutions must see to it that their salespersons do not churn, or make their customers frequently buy and sell securities, rather than recommending financial instruments that matches the knowledge, experience, assets and investment purposes of customers.
In the illustrative drawing below, it can be inferred that, based on the age, investment ratio, motive and other factors, the investor does not have ample reason to change the currently held financial instruments to something else. KIBIT, as an AI engine, learns such tacit knowledge, makes judgment similar or close to human judgment at a high speed, and finds similar records from among a massive volume of records.
03“Weight Refinement”- a special function of KIBIT
KIBIT is equipped with “Weight Refinement,” an algorithm unique to FRONTEO that optimizes the degree of importance attributed to words. Weight Refinement makes it possible for KIBIT to identify data characteristics that cannot be found through the calculation of transinformation alone. As its feature, the algorithm allows KIBIT to perform fully with a small volume of data and learn without building a large-scale server environment.
※Check out our “FRONTEO Research and Development Report”for more details about KIBIT.