Neural Network IT Solution Recommendation via Vector Embedding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI systems for recommending IT solutions lack effectiveness in providing comprehensive and secure recommendations, as they rely on limited training data and fail to incorporate both internal and external articles, leading to incomplete and insecure solution suggestions.
Innovation Solution
The development of a neural network-based system that extends training data by embedding internal and external IT solution articles into a unified vector space, allowing for the selection of relevant articles based on user queries and filtering them according to security credentials, thereby enhancing the recommendation accuracy and security.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the neural network is trained only on internal IT solution articles, then security is maintained, but the recommendation comprehensiveness and solution relevance deteriorate
Solution Approach 1:
The patent segments the training data into two distinct categories: internal IT solution articles (restricted access) and external IT solution articles (unrestricted access). The neural network is trained separately on these two segments, allowing it to learn from both sources while maintaining security boundaries. This segmentation enables the system to recommend from a broader article pool without compromising security, as the network learns to appropriately source recommendations based on the problem type and user context.
Solution Approach 2:
The patent introduces an intermediary component that acts as a bridge between internal and external articles. This intermediary layer processes and integrates information from both sources, allowing the neural network to leverage external knowledge while maintaining security protocols for internal content. The intermediary enables comprehensive recommendations by mediating between the security constraints and the need for broad knowledge access.
2Measurement precision
If the training data set is extended to include external articles, then solution relevance improves, but system complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-processing and embedding both internal and external articles into a unified vector space before training the neural network. This preprocessing step converts diverse article formats and sources into a standardized representation, simplifying the subsequent training process. By performing this preparation work in advance, the system avoids the complexity of handling heterogeneous data during runtime, thus improving solution relevance without proportionally increasing system complexity.
3Measurement precision
If both internal and external articles are integrated into the training set, then recommendation accuracy improves, but data processing complexity increases
Solution Approach 1:
The patent applies parameter changes by transforming the fundamental parameter of article representation into a unified vector space with consistent dimensionality. Both internal and external articles are converted to embeddings with the same parameter structure, allowing the neural network to process them uniformly. This parameter standardization enables accurate recommendations from diverse sources while minimizing data processing complexity, as the system handles all articles through the same processing pipeline.
Data Source
AI summary
Methods, systems, and apparatus for extended training data sets for neural network-based IT solutions. In one aspect, a method includes collecting, support data generated during multiple support events including a corpus of articles referenced during the support event. Within the corpus of articles, restricted and unrestricted articles referenced in the support data and restricted and unrestricted articles not referenced in the support data are identified. Embedded vectors are generated for each article that is referenced in the support data from the article and a subset of the support data that references the article, and for each article that is not referenced by the support data, an embedded vector from only the article. A dimensionality of the embedded vectors is reduced and a neural network is trained using the embedded vectors to select a particular article of the corpus of articles responsive to a new support event.