Back to results
US 2018/0018321 A1Published application
Avoiding sentiment model overfitting in a machine language model
Assigneesalesforce.com, inc.
InventorsMichael Jones, Scott Enman, Collin Chun-kit Lee, David Campbell, Christopher John Nicholls
Life of this patent
Estimating the term…
Abstract
Provided are techniques for avoiding sentiment model overfitting in a machine language model. A current list of keywords in a current sentiment model can be updated to create a proposed list of keywords in a proposed sentiment model. Machine-generated sentiment results, based on the proposed sentiment model, are presented to identify model overfitting, without revising the current set of keywords. The proposed set of keywords can be edited, and when overfitting is not present, the current list of keywords is replaced by the proposed list of keywords.
Claims18
A method, comprising:
- determining, using a first sentiment model having a current list of keywords and respective weights, respective sentiment results for a plurality of posts, wherein the determining is performed using a processing device;
- receiving, via a user interface, sentiment feedback for the respective sentiment results;
- displaying a first set of posts which are in the plurality of posts and have a sentiment feedback not matching the respective sentiment results;
- displaying:
- keywords which appear in the first set of posts;
- and
- the keywords' respective weights;
- receiving a first change including at least one of:
- a first proposed change to a respective weight of one or more of the keywords;
- a first proposed deletion of one or more of the keywords;
- or
- a first proposed new keyword and a respective weight of the first proposed new keyword;
- maintaining unchanged the current list of keywords and respective weights while creating a proposed list of keywords and respective weights by applying the first change to the current list of keywords and respective weights;
- analyzing the first set of posts, using the proposed list of keywords and respective weights;
- enabling identifying overfitting by displaying sentiment results for the first set of posts produced using the proposed list of keywords and respective weights;
- receiving a second change including at least one of:
- a second proposed change to a respective weight of one or more of the keywords;
- a second proposed deletion of one or more of the keywords;
- or
- a second proposed new keyword and a respective weight of the second proposed new keyword;
- updating the proposed list of keywords and respective weights by applying the second change to the proposed list of keywords and respective weights;
- and
- after applying the second change to the proposed list of keywords and respective weights, re-analyzing the first set of posts using the proposed list of keywords and respective weights;
- enabling identifying overfitting by displaying sentiment results for the first set of posts produced using the proposed list of keywords and respective weights;
- and
- creating a second sentiment model by replacing the current list of keywords and respective weights with the second proposed list of keywords and respective weights.
23 more elements
A system for mitigating sentiment model overfitting, the system comprising:
- a processor;
- and
- a memory communicably coupled to the processor, wherein the memory includes computer instructions configured to cause the processor to:
- determine, using a first sentiment model having a current list of keywords and respective weights, respective sentiment results for a plurality of posts;
- receive sentiment feedback for the respective sentiment results;
- display a first set of posts which are in the plurality of posts and have a sentiment feedback not matching the respective sentiment results;
- display:
- keywords which appear in the first set of posts;
- and
- the keywords' respective weights;
- receive a first change including at least one of:
- a first proposed change to a respective weight of one or more of the keywords;
- a first proposed deletion of one or more of the keywords;
- or
- a first proposed new keyword and a respective weight of the first proposed new keyword;
- maintain unchanged the current list of keywords and respective weights while creating a proposed list of keywords and respective weights by applying the first change to the current list of keywords and respective weights;
- analyze, using the proposed list of keywords and respective weights, the first set of posts;
- enable identifying overfitting by displaying sentiment results for the first set of posts produced using the proposed list of keywords and respective weights;
- receive a second change including at least one of:
- a second proposed change to a respective weight of one or more of the keywords;
- a second proposed deletion of one or more of the keywords;
- or
- a second proposed new keyword and a respective weight of the second proposed new keyword;
- update the proposed list of keywords and respective weights by applying the second change to the proposed list of keywords and respective weights;
- and
- after applying the second change to the proposed list of keywords and respective weights, re-analyze the first set of posts using the proposed list of keywords and respective weights;
- enable the identification of overfitting by the display of sentiment results for the first set of posts produced using the proposed list of keywords and respective weights;
- and
- create a second sentiment model by replacing the current list of keywords and respective weights with the proposed list of keywords and respective weights.
26 more elements
Dependencies are inferred from the claim text.