Description

PROBLEM BEING SOLVED

The invention addresses the need for complex event forecasting in data streams.  While existing systems can detect activity patterns upon event streams, the ability to forecast when such complex events might occur is lacking. This invention seeks to provide a method and system that allows users to define patterns for complex events and construct probabilistic models to forecast when these events are expected to happen in a stream of events.

METHOD USED

The method proposed by this invention involves receiving a pattern defining a complex event and an event stream with multiple events. A probabilistic model is then constructed to forecast the occurrence of the complex event within the stream. Key to this method is the use of an automaton derived from the pattern and a high-, variable-order Markov model to construct the probabilistic model. By utilizing these components, the system can accurately forecast complex events based on the received pattern, providing precise intervals within which the event is expected to occur. This unique approach allows for efficient and accurate forecasting of complex events in data streams.

ADVANTAGES

Compared to existing methods, this invention offers several advantages in complex event forecasting.  The use of a probabilistic model allows for a more nuanced understanding of event stream behavior, leading to more reliable forecasts. Additionally, the method can handle complex event definitions in a declarative manner, enabling users to define relationships among events in a stream more effectively. Overall, this invention represents a significant advancement in complex event forecasting, offering high accuracy and low latency in predicting the occurrence of complex events within data streams.

Description

PROBLEM BEING SOLVED

The invention addresses the challenge of automatically tagging electronic documents, particularly those containing financial data, which are a mix of structured tables and unstructured text. While tagging tables with XBRL (Extensive Business Reporting Language) can be done using templates, tagging text notes manually is time-consuming and error-prone. Existing methods for tagging documents with XBRL tags often deliver unsatisfactory results, especially for numerical or date values, highlighting the need for a more accurate and efficient tagging solution.

METHOD USED

The invention proposes a method for automatically tagging electronic documents using deep learning techniques. The process involves preprocessing the document by extracting text, replacing numbers or dates with special numeric and magnitude-preserving symbols, and tokenizing the text. A deep learning module then determines the appropriate tags for each token. This method allows for accurate tagging of numerical and date values in text notes, which has been a challenge for existing tagging methods. By leveraging deep learning and this new tokenization technique, the invention can efficiently assign tags to different pieces of information in the document, making the tagging process more reliable and precise.

ADVANTAGES

Compared to previous inventions, the current method offers several advantages. It streamlines the tagging process for electronic documents, especially those with a mix of structured and unstructured data, by automating the tagging of text notes containing numerical or date values. The deep learning approach enhances the accuracy of tagging, ensuring that the right tags are assigned based on context, while it also offers a computational time speedup since it alleviates any token overfragmentation. By reducing the need for manual intervention and improving the efficiency of tagging financial documents, the invention provides a more effective solution for annotating electronic documents with XBRL tags.

Description

PROBLEM BEING SOLVED

The invention aims to address the critical issue of drug-drug interactions (DDIs) that can lead to adverse effects, hospital admissions, and even death. Current methods for identifying interactions are complex, costly, and time-consuming, requiring clinical trials, laboratory tests, and expert knowledge. The need for a fast, reliable, and automated system to predict drug interactions is essential for patient safety and effective healthcare.

METHOD USED

The invention presents a method for estimating the likelihood of drug interactions by analyzing a variety of documents to extract relevant information about drug entities and their associations. This extracted information is then combined with supplementary data from structured sources to create a unified graph representation. Machine learning techniques are applied to analyze the graph and identify new edges, indicating potential interactions between drugs. By automating this process, the method can efficiently predict drug interactions without the need for extensive testing or human expertise.

ADVANTAGES

Compared to existing tools and methods, the invention offers several advantages in predicting drug interactions. By leveraging a combination of document analysis, structured data sources, and machine learning, the method can quickly and accurately identify potential interactions between drugs. This approach reduces the reliance on costly and time-consuming experimental procedures, making it a more efficient and cost-effective solution for clinicians and researchers. Ultimately, the invention provides a valuable tool for improving patient safety and healthcare outcomes by enabling the timely prediction and avoidance of harmful drug interactions.