Claims:We claim:
1. A method for determining classification of text, comprising:
receiving, by a text classification system (101), a text from a plurality of texts (212);
generating, by the text classification system (101), a first vector representation (213) of the text using first trained model parameters (219) of a first Long Short Term Memory (LSTM) unit 601 and a second vector representation (213) of the text using second trained model parameters (219) of a second LSTM unit (602), wherein the first and second trained model parameters (219) are obtained based on training of the text classification system (101) using a plurality of similar pair of texts and a plurality of dissimilar pair of texts from the plurality of texts (212);
combining, by the text classification system (101), the first vector representation (213) and the second vector representation (213) using a combiner operator (222) to obtain a combined vector representation (214), wherein the combiner operator (222) is selected from a plurality of combiner operators (222) based on the training using accuracy (221) of a classifier (104) of the text classification system (101); and
providing, by the text classification system (101), the combined vector representation (214) to the classifier (104) for determining a classification (215) of the text.

2. The method as claimed in claim 1 and further comprising:
cleaning, by the text classification system (101), the text to remove irrelevant data from the text.

3. The method as claimed in claim 1, wherein the training using the similar pair of texts and the dissimilar pair of texts comprises:
receiving, by the text classification system (101), a first text and a second text, wherein the first text and the second text are one of the plurality of similar pair of texts and one of the plurality of dissimilar pair of texts;
generating, by the text classification system (101), a first labeled text (216) for the first text and a second labeled text (216) for the second text based on a plurality of labels (217) retrieved from data source (105) associated with the text classification system (101);
generating, by the text classification system (101), a third vector representation (213) of the first text using first model parameters (218) of the first LSTM unit (601) and a fourth vector representation (218) of the second text using second model parameters (218) of the second LSTM unit (602); and
modifying, by the text classification system (101), the first model parameters (218) and the second model parameters (218) to obtain the first trained model parameters (219) and the second trained model parameters (219) respectively based on norm distance (220) associated between the third vector representation (213) and the fourth vector representation (213).

4. The method as claimed in claim 3, wherein the first labeled text (216) comprises the first text and one of the plurality of labels (217) associated with the first text and the second labeled text (216) comprises the second text and one of the plurality of labels (217) associated with the second text.

5. The method as claimed in claim 3, wherein modifying the first model parameters (218) and the second model parameters (218) comprises proximating exponential of the negative of norm distance (220) to value zero for the dissimilar pair of texts and proximating the exponential of the negative of norm distance (220) to value one for the similar pair of texts.

6. The method as claimed in claim 1, wherein selecting the combiner operator 222 comprises:
receiving, by the text classification system (101), each of the plurality of texts (212);
generating, by the text classification system (101), a fifth vector representation (213) for each of the plurality of texts (212) using the first trained model parameters (219) of the first LSTM unit 601 and a sixth vector representation (213) for each of the plurality of texts (212) using the second trained model parameters (219) of the second LSTM unit (602);
combining, by the text classification system, the fifth vector representation (213) and the sixth vector representation (213), for each of the plurality of texts (212), using each of the plurality of combiner operators (222) to obtain corresponding a plurality of combined vector representations (214);
providing, by the text classification system, each of the plurality of combined vector representations (214) to the classifier (104) to obtain classification (215) for each of the plurality of combined vector representations (214);
determining, by the text classification system, accuracy (221) for each of the classification (215); and
selecting, by the text classification system, the combiner operator (222), for each of the plurality of texts (212), from the plurality of combiner operators (222), based on the accuracy (221) of the classification (215) of the plurality of combined vector representations (214).

7. A text classification system (101) for determining classification of text, comprises:
a processor (106); and
a memory (109) communicatively coupled to the processor (106), wherein the memory (109) stores processor-executable instructions, which, on execution, cause the processor to:
receive a text from a plurality of texts (212);
generate a first vector representation (213) of the text using first trained model parameters (219) of a first Long Short Term Memory (LSTM) unit 601 and a second vector representation (213) of the text using second trained model parameters (219) of a second LSTM unit (602), wherein the first and second trained model parameters (219) are obtained based on training of the text classification system (101) using a plurality of similar pair of texts and a plurality of dissimilar pair of texts from the plurality of texts (212);
combine the first vector representation (213) and the second vector representation (213) using a combiner operator (222) to obtain a combined vector representation (214), wherein the combiner operator (222) is selected from a plurality of combiner operators (222) based on the training using accuracy (221) of a classifier (104) of the text classification system (101); and
provide the combined vector representation (214) to the classifier (104) for determining a classification (215) of the text.

8. The text classification system (101) as claimed in claim 7 and further comprises the processor to clean the text to remove irrelevant data from the text.

9. The text classification system (101) as claimed in claim 7, wherein the training using the similar pair of texts and the dissimilar pair of texts, comprises:
receiving a first text and a second text, wherein the first text and the second text are one of the plurality of similar pair of texts and one of the plurality of dissimilar pair of texts;
generating a first labeled text (216) for the first text and a second labeled text (216) for the second text based on a plurality of labels (217) retrieved from data source 105 associated with the text classification system (101);
generating a third vector representation (213) of the first text using first model parameters (218) of the first LSTM unit (601) and a fourth vector representation (218) of the second text using second model parameters (218) of the second LSTM unit (602); and
modifying the first model parameters (218) and the second model parameters (218) to obtain the first trained model parameters (219) and the second trained model parameters (219) respectively based on norm distance (220) associated between the third vector representation (213) and the fourth vector representation (213).

10. The text classification system (101) as claimed in claim 9, wherein the first labeled text (216) comprises the first text and one of the plurality of labels (217) associated with the first text and the second labeled text (216) comprises the second text and one of the plurality of labels (217) associated with the second text.

11. The text classification system (101) as claimed in claim 9, wherein modifying the first model parameters (218) and the second model parameters (218) comprises proximating exponential of the negative of norm distance (220) to value zero for the dissimilar pair of texts and proximating the exponential of the negative of norm distance (220) to value one for the similar pair of texts.

12. The text classification system (101) as claimed in claim7, wherein selecting the combiner operator 222 comprises:
receiving each of the plurality of texts (212);
generating a fifth vector representation (213) for each of the plurality of texts (212) using the first trained model parameters (219) of the first LSTM unit (601) and a sixth vector representation (213) for each of the plurality of texts (212) using the second trained model parameters (219) of the second LSTM unit (602);
combining, by the text classification system, the third vector representation (213) and the fourth vector representation (213) using each of the plurality of combiner operators (222) to obtain corresponding a plurality of combined vector representations (214);
providing, by the text classification system, the plurality of combined vector representations (214) to the classifier (104) to obtain classification (215) for each of the plurality of combined vector representations (214);
determining, by the text classification system, accuracy (221) for each of the classification (215); and
selecting, by the text classification system, the combiner operator (222) from the plurality of combiner operators (222), based on the accuracy (221) of the classification (215) of the plurality of combined vector representations (214).

Dated this 17th day of February 2017

R Ramya Rao
Of K&S Partners
Agent for the Applicant , Description:TECHNICAL FIELD
The present subject matter is related in general to the field of determining classification, more particularly, but not exclusively to a method and system for converting text to a vector representation for determining classification of the text.