Regularized Character Embeddings for Entity Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional natural language processing systems for entity matching are inefficient in processing infrequent or out-of-domain character sequences, as they rely on observing frequently occurring character or word sequences in datasets, leading to decreased performance in matching rare input sequences or misspelled names.
Innovation Solution
The system encodes input and candidate names using a regularized character set with fewer characters than the natural language character set, employing a character encoder, character embedding component, and word embedding component to produce regularized sequences and vectors, which are then used to compute a similarity score for matching, utilizing deep learning networks like LSTM or transformer models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional NLP networks rely on frequently observed character or word sequences in datasets, then processing efficiency for common sequences is improved, but performance deteriorates when processing infrequent or out-of-domain character sequences
Solution Approach 1:
The patent transforms the input character sequences by applying parameter changes - specifically, it maps characters to numerical indices based on a predefined vocabulary, converting textual data into a numerical format that the neural network can process efficiently while maintaining the semantic relationships between characters
Solution Approach 2:
The patent introduces an intermediary embedding layer that acts as a bridge between the input character sequences and the neural network processing. This embedding layer transforms discrete character indices into continuous vector representations, allowing the network to handle both frequent and infrequent sequences effectively by learning semantic relationships in the embedding space
2Device complexity
If the character set size is reduced to a regularized set with fewer characters, then the complexity of processing is reduced and generalization improves, but information loss may occur
Solution Approach 1:
The patent creates a simplified copy of the character set - a regularized vocabulary containing only the most relevant characters. By mapping input sequences to this reduced vocabulary and using embedding layers to preserve semantic information, the system achieves lower processing complexity without significant information loss, as the embedding vectors capture the essential meaning of characters
Data Source
AI summary
Systems and methods for natural language processing are described. One or more embodiments of the disclosure provide an entity matching apparatus trained using machine learning techniques to determine whether a query name corresponds to a candidate name based on a similarity score. In some examples, the query name and the candidate name are encoded using a character encoder to produce a regularized input sequence and a regularized candidate sequence, respectively. The regularized input sequence and the regularized candidate sequence are formed from a regularized character set having fewer characters than a natural language character set.