System, method and apparatus for automatic categorization and assessment of billing narratives
A system for automatic categorization and assessment of billing narratives has a semantic engine that classifies billing entries with descriptions expressed in natural language into standardized categories of activity and task objective. The classification is by machine learning methods via training data that is maintained, updated and extended as needed. A rules engine applies rules to the categorized invoice data to analyze the data, report violations to a user/consumer of billed services and to perform related calculations.
1. A system for the automated analysis of actual invoices and timekeeper narratives included therein, comprising:
a computer having a processor;
a data store accessible to the computer and capable of storing invoice data in digital textural form;
a semantic engine capable of running on the processor, the semantic engine having a learning layer and a processing layer, the learning layer being configured to receive training data containing accurately decomposed and categorized sample invoice data, including narratives, from a training sample of invoice data and develop a training model, the decomposed and categorized sample invoice data including a set of standardized categories of actions, each of which describes an action that may be performed by a timekeeper, and objects of actions, each of which describes an object upon which an action is performed by a timekeeper, the processing layer being configured to process actual invoice data associated with an actual invoice in accordance with the training model, to decompose timekeeper narratives in the actual invoice data describing work performed by a service provider into fragments pertaining to actions performed and objects of actions by the service provider, and to categorize the decomposed fragments into at least one of the set of standardized categories of actions and objects of actions for forming categorized invoice data in accordance with the training model, and
a rules engine capable of running on the processor, the rules engine having access to at least one rule applicable to the at least one of the set of standardized categories of actions and objects of actions, the at least one rule including one or more billing guidelines,
wherein the rules engine is configured to receive the categorized invoice data from the semantic engine, to apply at least one of the one or more billing guidelines of the at least one rule to the categorized invoice data associated with at least one of the timekeeper narratives, and to generate an output consistent with the at least one rule, the output including results of an analysis including at least one determination of whether work associated with at least one of the timekeeper narratives has been optimally performed.
2. The system of claim 1 , further comprising a reporting engine, the reporting engine being configured to generate reports to a user of the categorized invoice data based on an application of the at least one rule, and
at least one additional computer connected to the computer via a network, the at least one additional computer being configured to permit the distribution of data storage and to process tasks performed by the system over the computer and the at least one additional computer.
3. The system of claim 1 , wherein the learning layer is configured to generate a database of pre-categorized text fragments based upon prior evaluation of the training sample of invoice data and to store the database on the data store accessible to the computer, and wherein the processing layer is configured to access the database.
4. The system of claim 3 , wherein the processing layer is further configured to evaluate an accuracy of the categorized invoice data, resulting in additional entries to the database.
5. The system of claim 1 , further comprising a database of a plurality of identities corresponding to a plurality of persons associated with at least one of the service provider or a recipient, and including a plurality of roles of the plurality of persons, to store the database of the plurality of persons in the data store, the semantic engine being configured to refer to the database of the plurality of persons and to incorporate the plurality of roles of the plurality of persons while processing the actual invoice data, including narratives.
6. The system of claim 1 , wherein at least one rule includes at least one of best practice norms or terms of engagement.
7. The system of claim 1 , wherein the at least one rule is configured to be changed under the direction of a recipient of the output.
8. The system of claim 1 , wherein the rules engine is configured to generate a pre-determined output message in response to detecting a violation of the at least one rule.
9. The system of claim 1 , wherein the processing layer of the semantic engine is configured to pre-process the actual invoice data by tokenizing the actual invoice data into word and sentence fragments and then transforming the word and sentence fragments into a lemmatized form.
10. The system of claim 9 , wherein the processing layer is configured to process the lemmatized form of word and sentence fragments by n gram extraction as part of a speech tagging and dependency parsing through analysis of syntactic structure of the actual invoice data.
11. The system of claim 9 , wherein named entities are extracted from the actual invoice data.
12. The system of claim 1 , wherein the learning layer is configured to implement model training, model aggregation and model evaluation.
13. The system of claim 12 , wherein the model training includes weighting, category binarization, feature selection and parameter tuning.
14. The system of claim 12 , wherein the model aggregation includes ensemble tuning, probability thresholding and result combination.
15. The system of claim 12 , wherein the model evaluation includes metric generation and error modeling.
16. The system of claim 1 , wherein the semantic engine is configured to develop a database of people and billing rates and to store the database of people and billing rates in the data store accessible to the computer, and is further configured to calculate charges by persons on behalf of the service provider.
17. The system of claim 1 , wherein the semantic engine is configured to match the name of at least one person named in the timekeeper narratives against a pre-determined set of names of individuals and to determine the categorization of the decomposed fragments into the at least one of the set of standardized categories of actions and objects of actions.
18. The system of claim 17 , wherein the pre-determined set of names of individuals include affiliation and titles of the individuals, the semantic engine being configured to determine whether the categorization of the decomposed fragments into the at least one of the set of standardized categories of actions and objects of actions should be classified as internal or external of the service provider.