Method and system for text classification based on learning of transferable feature representations from a source domain

A system for learning transferable feature representations, the system comprising one or more processors in a data processing server configured to:

• receive real-time input data from a computing device over a communication network, wherein the received input data comprises labeled instances of a source domain and unlabeled instances of a target domain;
• determine a source specific representation corresponding to the source domain and a common representation shared between the source domain and the target domain;
• classify the labeled instances of the source domain into a plurality of categories based on a first set of one or more representations, using a neural network, the neural network comprising:
• an input layer,
• a hidden layer connected to the input layer, the hidden layer comprising a plurality of sections, the plurality of sections comprising two or more of a source specific representation, a source representation, a global representation, wherein the hidden layer comprises the first set of one or more representations and the first set of one or more representations comprises the source specific representation and the source representation, the source representation comprising features that are common between a source-target pair in a plurality of source-target pairs from the source and target domains, and
• an output layer connected to the hidden layer;
• train a generalized classifier based on a positive contribution from the common representation using a domain regressor and the neural network, wherein the hidden layer comprises a second set of one or more representations and the second set of one or more representations comprises the source specific representation and the source representation;
• and
• automatically perform classification on the unlabeled instances of the target domain based on the trained generalized classifier.

A computer program product for use with a computer, the computer program product comprising a non-transitory computer readable medium, wherein the non-transitory computer readable medium stores a computer program code for learning transferable feature representations, wherein the computer program code is executable by one or more processors in a data processing server to:

• receive real-time input data from a computing device over a communication network, wherein the received input data comprises labeled instances of a source domain and unlabeled instances of a target domain;
• determine a source specific representation corresponding to the source domain, and a common representation shared between the source domain and the target domain;
• classify the labeled instances of the source domain based on a first set of one or more representations, using a neural network, the neural network comprising:
• an input layer,
• a hidden layer connected to the input layer, the hidden layer comprising a plurality of sections, the plurality of sections comprising two or more of a source specific representation, a source representation, a global representation, wherein the hidden layer comprises the first set of one or more representations and the first set of one or more representations comprises the source specific representation and the source representation, the source representation comprising features that are common between a source-target pair in a plurality of source-target pairs from the source and target domains;
• and
• an output layer connected to the hidden layer;
• train a generalized classifier based on a positive contribution from the common representation using a domain regressor and the neural network, wherein the hidden layer comprises a second set of one or more representations and the second set of one or more representations comprises the source specific representation and the source representation;
• and
• automatically perform classification on the unlabeled instances of the target domain based on the trained generalized classifier.