TEXT CATEGORIZATION WITH KNOWLEDGE TRANSFER FROM HETEROGENEOUS DATASETS

• US 20090171956A1
• Filed: 10/10/2008
• Published: 07/02/2009
• Est. Priority Date: 10/11/2007
• Status: Active Grant

• Current Assignee: Honda Motor Co., Ltd. (Honda Motor Company)
• Original Assignee: Honda Motor Co., Ltd. (Honda Motor Company)
• Inventors: Ratinov, Lev, Gupta, Rakesh
• Granted Patent: US 8,103,671 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]

Abstract

The present invention provides a method for incorporating features from heterogeneous auxiliary datasets into input text data for use in classification, a plurality of heterogeneous auxiliary datasets, such as labeled datasets and unlabeled datasets, are accessed after receiving input text data. A plurality of features are extracted from each of the plurality of heterogeneous auxiliary datasets. The plurality of features are combined with the input text data to generate a set of features which may potentially be used to classify the input text data. Classification features are then extracted from the set of features and used to classify the input text data. In one embodiment, the classification features are extracted by calculating a mutual information value associated with each feature in the set of features and identifying features having a mutual information value exceeding a threshold value.

Claims

1. A computer-based method for classifying input text data by obtaining information from multiple datasets, the method comprising the steps of:

receiving the input text data;

accessing a plurality of heterogeneous datasets, the plurality of heterogeneous datasets each including text data;

generating a set features from the plurality of heterogeneous datasets, the set of features including one or more features from each of the plurality of heterogeneous datasets;

selecting one or more classification features from the set of features; and

generating an augmented input text data by combining the input text data and the one or more classification features; and

applying a classifier to the augmented input text data to associate the input text data with a category.

2. The computer-implemented method of claim 1, wherein the plurality of heterogeneous datasets comprises a labeled dataset and an unlabeled dataset.

3. The computer-implemented method of claim 1, wherein generating the set of features from the plurality of heterogeneous datasets comprises:

determining whether a dataset is a labeled dataset or an unlabeled dataset;

responsive to determining the dataset is a labeled dataset, generating a vector representation of a category of the dataset;

generating a reduced vector representation by truncating the vector representation of the category of the dataset;

determining a similarity value between the reduced vector representation and the input text data; and

responsive to the similarity value exceeding a threshold value, including the reduced vector representation in the set of features.

4. The computer-implemented method of claim 3, wherein the vector representation of the category of the dataset comprises a term-frequency-inverse document frequency vector.

5. The computer-implemented method of claim 3, wherein generating the reduced vector representation comprises:

selecting a centroid of the vector representation of the category of the dataset; and

truncating the centroid of the vector representation of the category of the dataset to a predefined number of words.

6. The computer-implemented method of claim 3, wherein determining the similarity value between the reduced vector representation and the input text data comprises:

computing a cosine similarity between the reduced vector representation and the input text data.

7. The computer-implemented method of claim 1, wherein generating the set of features from the plurality of heterogeneous datasets comprises:

determining whether a dataset is a labeled dataset or an unlabeled dataset;

responsive to determining the dataset is an unlabeled dataset, generating a plurality of clusters from the dataset;

generating a vector representation of a cluster;

generating a subset of the vector representation of the cluster;

determining a similarity value between the subset of the vector representation and the input text data; and

responsive to the similarity value exceeding a threshold value, including the reduced vector representation in the set of features.

8. The computer-implemented method of claim 7, wherein generating the plurality of clusters from the dataset comprises:

generating a defined number of clusters using unsupervised agglomerative clustering.

9. The computer-implemented method of claim 7, wherein the vector representation of the cluster comprises a term-frequency-inverse document frequency vector.

10. The computer-implemented method of claim 7, wherein generating the subset of the vector representation of the cluster comprises:

selecting a predetermined number of words from the vector representation of the cluster.

11. The computer-implemented method of claim 7, wherein determining the similarity value between the subset of the vector representation and the input text data comprises:

computing a cosine similarity between the subset of the vector representation and the input text data.

12. The computer-implemented method of claim 1, wherein selecting one or more classification features from the set of features comprises:

determining a mutual information score associated with a feature from the set of features and the input text data; and

responsive to the mutual information score exceeding a threshold, including the feature in the set of features.

13. A computer program product, comprising a computer readable storage medium storing computer executable code classifying input text data by obtaining information from multiple datasets, the computer executable code performing the steps of:

receiving the input text data;

accessing a plurality of heterogeneous datasets, the plurality of heterogeneous datasets each including text data;

generating a set features from the plurality of heterogeneous datasets, the set of features including one or more features from each of the plurality of heterogeneous datasets;

selecting one or more classification features from the set of features;

generating an augmented input text data by combining the input text data and the one or more classification features; and

applying a classifier to the augmented input text data to associate the input text data with a category.

14. The computer program product of claim 13, wherein the plurality of heterogeneous datasets comprises a labeled dataset and an unlabeled dataset.

15. The computer program product of claim 13, wherein generating the set of features from the plurality of heterogeneous datasets comprises:

determining whether a dataset is a labeled dataset or an unlabeled dataset;

responsive to determining the dataset is a labeled dataset, generating a vector representation of a category of the dataset;

generating a reduced vector representation by truncating the vector representation of the category of the dataset;

determining a similarity value between the reduced vector representation and the input text data; and

responsive to the similarity value exceeding a threshold value, including the reduced vector representation in the set of features.

16. The computer program product of claim 15, wherein the vector representation of the category of the dataset comprises a term-frequency-inverse document frequency vector.

17. The computer program product of claim 15, wherein generating the reduced vector representation comprises:

selecting a centroid of the vector representation of the category of the dataset; and

truncating the centroid of the vector representation of the category of the dataset to a predefined number of words.

18. The computer program product of claim 15, wherein determining the similarity value between the reduced vector representation and the input text data comprises:

computing a cosine similarity between the reduced vector representation and the input text data.

19. The computer program product of claim 13, wherein generating the set of features from the plurality of heterogeneous datasets comprises:

determining whether a dataset is a labeled dataset or an unlabeled dataset;

responsive to determining the dataset is an unlabeled dataset, generating a plurality of clusters from the dataset;

generating a vector representation of a cluster;

generating a subset of the vector representation of the cluster;

determining a similarity value between the subset of the vector representation and the input text data; and

responsive to the similarity value exceeding a threshold value, including the reduced vector representation in the set of features.

20. The computer program product of claim 19, wherein generating the plurality of clusters from the dataset comprises:

generating a defined number of clusters using unsupervised agglomerative clustering.

21. The computer program product of claim 19, wherein the vector representation of the cluster comprises a term-frequency-inverse document frequency vector.

22. The computer program product of claim 19, wherein generating the subset of the vector representation of the cluster comprises:

selecting a predetermined number of words from the vector representation of the cluster.

23. The computer program product of claim 19, wherein determining the similarity value between the subset of the vector representation and the input text data comprises:

computing a cosine similarity between the subset of the vector representation and the input text data.

24. The computer program product of claim 13, wherein selecting one or more classification features from the set of features comprises:

determining a mutual information score associated with a feature from the set of features and the input text data; and

responsive to the mutual information score exceeding a threshold, including the feature in the set of features.