Language Model Sentence Recommendation Using Contextual History
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing electronic devices are inconvenient as they only recommend one word at a time, increasing the number of selections required for typing, and often fail to suggest appropriate completions for user-intended sentences based on input words.
Innovation Solution
An electronic apparatus that uses a machine-trained language model, specifically a recurrent neural network (RNN), to recommend completed sentences by applying contextual information from previously transmitted and received messages, incorporating the inputted text to predict and suggest intended sentences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the electronic apparatus recommends one word at a time based on user input, then the typing effort is reduced, but the number of selections required increases
Solution Approach 1:
The patent segments the sentence completion recommendation into multiple candidate words ranked by probability, allowing the user to select from a curated list rather than typing the entire sentence. This segmentation approach reduces typing effort while limiting selection time through prioritized candidates.
Solution Approach 2:
The system performs preliminary action by pre-calculating and ranking multiple candidate completion words before the user needs to type or select. The language model generates probable continuations in advance, so when the user pauses or requests suggestions, ready-made options are immediately available, reducing both typing and selection time.
2Device complexity
If the electronic apparatus recommends words based only on the input word, then the recommendation process is simple, but appropriate sentence completions are often not suggested
Solution Approach 1:
The patent implements feedback by using the user's partial input as a query to the language model, which then returns ranked candidate completions. The system continuously refines recommendations based on the user's input progress and selection behavior, improving accuracy without significantly increasing complexity through iterative feedback loops.
Solution Approach 2:
The language model acts as an intermediary between the user's partial input and the suggested completions. Rather than using simple rule-based matching, the neural language model processes the input context and generates probabilistic predictions, serving as a sophisticated mediator that improves recommendation accuracy while keeping the user interface simple.
3Measurement precision
If the electronic apparatus uses a machine trained language model with contextual information, then the accuracy of recommending intended sentences is improved, but the processing complexity increases
Solution Approach 1:
The patent applies partial action by using a pre-trained language model that has already learned from vast amounts of data, rather than training a model from scratch. The system leverages the pre-existing knowledge in the model to generate accurate completions with minimal additional processing, achieving high accuracy without the excessive complexity of full model training and customization.
Data Source
AI summary
Disclosed is an electronic apparatus. The electronic apparatus includes a communicator comprising communication circuitry; and a processor configured to control the electronic apparatus to: obtain a recommended message including the inputted text using a machine trained language model, based on a text being input, and control the communicator to transmit the recommended message to an external electronic apparatus based on receiving an input for transmitting the recommended message. The processor may be configured to control the electronic apparatus to obtain the recommended message by applying the inputted text and one or more message transmitted and received between the communicator and the external electronic apparatus prior to the text being input to the language model.