Predicting Binary Outcomes Of An Activity
Assigneesalesforce.com, inc.
InventorPaul Walter Hubenig
Life of this patent
Estimating the term…
Abstract
A binary outcome of an activity is predicted based on samples of the activity. The activity is characterized by features that can take on any of a set of mutually exclusive levels. An initial candidate pool of terms is selected. The terms are feature levels or combinations of feature levels. The candidate terms are used to create two ordered pools of terms, one including terms that are positively predictive and another including terms that are negatively predictive. The terms in each pool are ordered by strength of predictiveness and diversity of predictiveness relative to terms higher in the order. A final set of terms is selected by combining terms from these two pools.
Claims23
A method implemented on a computer system for predicting a binary outcome of an activity based on samples of the activity, the activity characterized by features that can take on any of a set of mutually exclusive levels for that feature and each sample of the activity including the feature levels and the binary outcome for that sample, the method comprising:
- selecting an initial candidate pool of terms for the activity, wherein each term is a feature level or a combination of feature levels from different features;
- based on the samples of the activity:
- creating a positive ordered pool of terms and a negative ordered pool of terms from the initial candidate pool of terms, the positive ordered pool containing terms that are positively predictive of the binary outcome and the negative ordered pool containing terms that are negatively predictive of the binary outcome;
- and
- selecting a final set of terms from the terms in the positive ordered pool and the terms in the negative ordered pool;
- and
- predicting the binary outcome of future instances of the activity based on levels of the terms in the final set of terms.
A method implemented on a computer system for predicting a binary outcome of an activity based on samples of the activity, the activity characterized by features that can take on any of a set of mutually exclusive levels for that feature and each sample of the activity including the feature levels and the binary outcome for that sample, the method comprising:
- selecting an initial candidate pool of terms for the activity, wherein each term is a feature level or a combination of feature levels from different features;
- based on the samples of the activity, calculating a partial mutual information for the terms in the initial candidate pool;
- selecting a subset of terms from the initial candidate pool with strongest partial mutual information;
- placing terms from the subset of terms into either a group of positively predictive terms or a group of negatively predictive terms;
- for the group of positively predictive terms and the group of negatively predictive terms, creating a positive ordered pool of terms and a negative ordered pool of terms by, based on the samples of the activity:
- calculating a joint partial mutual information for pairs of terms within that group;
- and
- ordering the terms within that group based on the joint partial mutual information;
- selecting a term that is highest in either the positive or the negative ordered pool as a first term in a final set of terms;
- successively adding more terms to the final set by selecting between the next highest unselected term in the positive ordered pool and the next highest unselected term in the negative ordered pool;
- and
- predicting the binary outcome of future instances of the activity based on levels of the terms in the final set of terms.
A non-transitory computer-readable storage medium storing executable computer program instructions for predicting a binary outcome of an activity based on samples of the activity, the activity characterized by features that can take on any of a set of mutually exclusive levels for that feature and each sample of the activity including the feature levels and the binary outcome for that sample, the instructions executable by a processor and causing the processor to perform a method comprising:
- selecting an initial candidate pool of terms for the activity, wherein each term is a feature level or a combination of feature levels from different features;
- based on the samples of the activity:
- creating a positive ordered pool of terms and a negative ordered pool of terms from the initial candidate pool of terms, the positive ordered pool containing terms that are positively predictive of the binary outcome and that are ordered by strength of predictiveness and diversity of predictiveness relative to terms higher in the order, and the negative ordered pool containing terms that are negatively predictive of the binary outcome and that are ordered by strength of predictiveness and diversity of predictiveness relative to terms higher in the order;
- and
- selecting a final set of terms by selecting a term that is highest in either the positive or the negative ordered pool, and successively adding more terms by selecting between the next highest unselected term in the positive ordered pool and the next highest unselected term in the negative ordered pool;
- and
- predicting the binary outcome of future instances of the activity based on levels of the terms in the final set of terms.
A method implemented on a computer system for predicting a binary outcome of an activity based on samples of the activity, the activity characterized by features that can take on any of a set of mutually exclusive levels for that feature and each sample of the activity including the feature levels and the binary outcome for that sample, the method comprising:
- selecting an initial candidate pool of terms for the activity, wherein each term is a feature level or a combination of feature levels from different features;
- based on the samples of the activity, calculating a partial mutual information for the terms in the initial candidate pool;
- selecting a subset of terms from the initial candidate pool with strongest partial mutual information;
- selecting a final set of terms from the subset of terms;
- and
- predicting the binary outcome of future instances of the activity based on levels of the terms in the final set of terms.
A method implemented on a computer system for predicting a binary outcome of an activity based on samples of the activity, the activity characterized by features that can take on any of a set of mutually exclusive levels for that feature and each sample of the activity including the feature levels and the binary outcome for that sample, the method comprising:
- ordering terms in a group of terms for the activity, wherein each term is a feature level or a combination of feature levels from different features, comprising:
- based on the samples of the activity, calculating a joint partial mutual information for pairs of terms within the group;
- and
- ordering the terms based on the joint partial mutual information;
- selecting a final set of terms that includes terms from the ordered group of terms;
- and
- predicting the binary outcome of future instances of the activity based on levels of the terms in the final set of terms.
Dependencies are inferred from the claim text.