A personalization algorithm that analyzes users’ potential interests from textual descriptions.
Analzes users’ qualitative taste preferences (e.g. avant-garde, pop, luxury, modern, etc.) from texts written by the user with minimal computational overhead, and achieving accurate personalization for diverse tastes.
Leveraging the geometric properties of a linguistic hyperspace and our proprietary mathematical machine-learning algorithm, we can analyze a user’s preferences and personality traits described in text without relying on accumulated behavioral data. This is achieved through a unique semantic hyperplane algorithm in the vocabulary embedding space, which defines conceptual groups by antonyms, and operates within minimal computation time. As a result, we infer subtle inclinations of users who have scant behavioral histories or limited initial information, thus enhancing the service’s product suggestion, recommendation, and personalization capabilities.
(Patent granted: Patent No. 7393772)
It also solves the cold start problem caused by insufficient accumulation of initial behavioral data, a longstanding challenge in conventional recommendation systems.
In conventional recommendation algorithms, they do not work until enough behavioral data has been accumulated, rendering them powerless for new users who are actually critical. This issue is known as the cold start problem (where systems cannot operate without accumulated data), representing a fatal flaw in traditional algorithms. In contrast, the LSH algorithm does not require the accumulation of behavioral data. Even a brief descriptive text from the user is enough to instantly infer their interests. Consequently, it can immediately provide personalization for new users at the crucial initial stage.
In this post–mass media era, it can precisely address the increasingly diverse and niche interests of modern audiences.
Traditional behavior data-based recommendation algorithms have ensured their accuracy by consistently proposing products popular among the majority eventually, which has led to criticism that they do not effectively cater to diverse tastes and preferences.
Meanwhile, in today's environment where mass media is gradually disappearing, there is a growing need for personalization algorithms that can handle increasingly diversified interests.
By leveraging its unique linguistic hyper-spatial geometry, the LSH algorithm can arbitrarily define and capture niche interest vectors, such as liking simple things, liking unusual things, liking flashy things, liking practical things, or liking things from the 60s.
English