Method for predicting business income from user transaction data

1. A computer implemented method, comprising:

generating a vector of features from data related to a historical transaction;

generating a probability that the historical transaction belongs to a specific classification with respect to income;

training a multinomial classifier using the vector, a label associated with the historical transaction, and the probability;

assigning weights to a sparse matrix made up of a plurality of vectors that include the vector to train the multinomial classifier;

obtaining data related to a new transaction from a data stream for an account of an online service;

splitting the data related to the new transaction into a set of unigrams;

generating a new vector of features from the data related to the new transaction, the new vector including a set of values that correspond and are assigned to the set of unigrams;

determining a classification with respect to income for the new transaction by applying the multinomial classifier to the new vector;

labeling the new transaction with the classification;

presenting the classification to a view of a graphical user interface; and

populating, using the classification, one or more fields of a form that is maintained by the online service.

2. The method of claim 1 , further comprising:

receiving a reclassification of the new transaction;

populating, using the reclassification instead of the classification, one or more fields of the form that is maintained by the online service; and

updating, using the reclassification, the multinomial classifier.

3. The method of claim 1 ,

wherein at least one of the features has been filtered using a custom stop-word dictionary developed through empirical testing of the multinomial classifier, and

wherein the classification is based on a probability generated by applying the multinomial classifier to the new vector.

4. The method of claim 1 , wherein one of the features identifies a week day on which the historical transaction occurred.

5. The method of claim 1 , further comprising:

applying a threshold based on a precision-recall curve to the probability when determining the classification with respect to income for the new transaction.

6. The method of claim 1 , further comprising the operation of:

retraining the multinomial classifier at an end of a predetermined period using transactions which occurred during the predetermined period.

7. The method of claim 1 , wherein the online service is a massively multi-user online service.

8. The computer implemented method of claim 1 , further comprising:

obtaining data related to a plurality of historical transactions,

wherein each historical transaction is associated with a second account provided by the online service and with a label based on a click stream of graphical user interface interactions, and

wherein the label identifies the historical transaction as belonging to a specific classification with respect to income.

9. The method of claim 8 , further comprising:

obtaining data related to tax filing for a plurality of accounts of the online service;

mining the data related to tax filing and the historical transactions using a clustering technique to identify potential sources of income; and

training the multinomial classifier to generate a probability that a specific transaction is one of the potential sources of income.

10. A non-transitory computer-readable storage medium storing instructions, which when executed, perform operations as follows:

generate a vector of features from the data related to each historical transaction;

generate a probability that the historical transaction belongs to a specific classification with respect to income;

train a multinomial classifier using the vector, a label associated with the historical transaction, and the probability;

assign weights to a sparse matrix made up of a plurality of vectors that include the vector to train the multinomial classifier;

obtain data related to a new transaction from a data stream for an account of an online service;

split the data related to the new transaction into a set of unigrams;

generate a new vector of features from the data related to the new transaction, the new vector including a set of values that correspond and are assigned to the set of unigrams;

determine a classification with respect to income for the new transaction by applying the multinomial classifier to the new vector;

label the new transaction with the classification;

present the classification to a graphical user interface; and

populate, using the classification, one or more fields of a form that is maintained by the online service.

11. The non-transitory computer-readable storage medium of claim 10 , further comprising instructions to:

receive a reclassification of the new transaction;

populate, using the reclassification instead of the classification, one or more fields of the form that is maintained by the online service; and

update, using the reclassification, the multinomial classifier.

12. The non-transitory computer-readable storage medium of claim 10 ,

wherein at least one of the features has been filtered using a custom stop-word dictionary developed through empirical testing of the multinomial classifier, and

wherein the classification is based on a probability generated by applying the multinomial classifier to the new vector.

13. The non-transitory computer-readable storage medium of claim 10 , wherein one of the features identifies a week day on which the historical transaction occurred.

14. The non-transitory computer-readable storage medium of claim 10 , further comprising instructions to:

apply a threshold based on a precision-recall curve to a probability when determining the classification with respect to income for the new transaction.

15. The non-transitory computer-readable storage medium of claim 10 , further comprising instructions to:

retrain the multinomial classifier at an end of a predetermined period using transactions which occurred during the predetermined period.

16. The non-transitory computer-readable storage medium of claim 10 , wherein the online service is a massively multi-user online service.

17. The non-transitory computer-readable storage medium of claim 10 , further comprising instructions to:

obtain data related to a plurality of historical transactions,

wherein each historical transaction is associated with a second account provided by the online service and with a label based on a click stream of graphical user interface interactions, and

wherein the label identifies the historical transaction as belonging to a specific classification with respect to income.

18. The non-transitory computer-readable storage medium of claim 17 , further comprising instructions to:

obtain data related to tax filing for a plurality of accounts of the online service;

mine the data related to tax filing and the historical transactions using a clustering technique to identify potential sources of income; and

train the multinomial classifier to generate the probability that a specific transaction is one of the potential sources of income.

19. A system comprising:

a processor;

a storage storing instructions which, when executed by the processor, perform operations as follows:

generate a vector of features from the data related to each historical transaction;

generate a probability that the historical transaction belongs to a specific classification with respect to income;

train a multinomial classifier using the vector, a label associated with the historical transaction, and the probability;

assign weights to a sparse matrix made up of a plurality of vectors that include the vector to train the multinomial classifier;

obtain data related to a new transaction from a data stream for an account of an online service;

split the data related to the new transaction into a set of unigrams;

generate a new vector of features from the data related to the new transaction, the new vector including a set of values that correspond and are assigned to the set of unigrams;

determine a classification with respect to income for the new transaction by applying the multinomial classifier to the new vector;

label the new transaction with the classification;

present the classification to a graphical user interface; and

populate, using the classification, one or more fields of a form that is maintained by the online service.

20. The system of claim 19 , wherein the instructions further perform operations as follows:

receive a reclassification of the new transaction;

populate, using the reclassification instead of the classification, one or more fields of the form that is maintained by the online service; and

update, using the reclassification, the multinomial classifier.