# US 2020/0004823 A1 · Method and device for extracting point of interest from natural language sentences

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2020/0004823 A1
- Document ID: US20200004823A1
- Status: Published application
- Kind: A1
- Application number: 16/111,273
- Filed: 2018-08-24
- Published: 2020-01-02
- Earliest claimed: 2018-06-30
- Assignees: Wipro Limited
- Applicants: Wipro Limited
- Inventors: Arindam CHATTERJEE; Kartik SUBODH BALLAL
- Web page: https://patentgrep.com/patent/US20200004823A1

## CPC

- G06F 40/279: Recognition of textual entities
- G06F 40/30: Semantic analysis
- G06F 40/205: Parsing
- G06F 40/253: Grammatical analysis; Style critique
- G06F 40/289: Phrasal analysis, e.g. finite state techniques or chunking
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06N 5/022: Knowledge engineering; Knowledge acquisition
- G06N 3/042: Knowledge-based neural networks; Logical representations of neural networks
- G06N 5/04: Inference or reasoning models
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A method and device for extracting Point of Interest (POI) from natural language sentences is disclosed. The method includes creating an input vector comprising a plurality of parameters for each target word in a sentence inputted by a user. The method further includes processing for each target word, the input vector through a trained bidirectional LSTM neural network, which is trained to identify POI from a plurality of sentences. The method includes associating POI tags to each target word in the sentence based on processing of associated input vector through the trained bidirectional LSTM neural network. The method further includes extracting POI text from the sentence based on the POI tags associated with each target word in the sentence. The method further includes providing a response to the sentence inputted by the user based on the POI text extracted from the sentence.

## Claims

### Claim 1

A method of extracting Point of Interest (POI) from natural language sentences, the method comprising:
creating, by a POI processing device, an input vector comprising a plurality of parameters for each target word in a sentence inputted by a user, wherein the plurality of parameters for each target word comprises a Part of Speech (POS) vector associated with a target word and at least two words preceding the target word, a word embedding for the target word, a word embedding for a head word of the target word in the dependency parse tree of the input sentence, and a dependency label for the target word;
processing for each target word, by the POI processing device, the input vector through a trained bidirectional Long Short Term Memory (LSTM) neural network;
assigning, by the POI processing device, POI tags to each target word in the sentence based on processing of associated input vector through the trained bidirectional LSTM neural network;
extracting, by the POI processing device, POI text from the sentence based on the POI tags associated with each target word in the sentence; and
providing, by the POI processing device, a response to the sentence inputted by the user based on the POI text extracted from the sentence.

### Claim 2 (depends on 1)

The method of claim 1 further comprising determining the plurality of parameters for each target word in the sentence inputted by the user.

### Claim 3 (depends on 2)

The method of claim 2, wherein the response comprises at least one of an answer to a user query and an action corresponding to the user query.

### Claim 4 (depends on 1)

The method of claim 1, wherein the dependency label for the target word indicates a relation of the target word with the head word in the sentence.

### Claim 5 (depends on 1)

The method of claim 1 further comprising:
training a bidirectional LSTM neural network to identify the POI tags for words within sentences.

### Claim 6 (depends on 5)

The method of claim 5, wherein training the bidirectional LSTM neural network comprises:
assigning the POI tags to each word in natural language sentences retrieved from a data repository of natural language sentences comprising a plurality of POI scenarios;
inputting, iteratively, assigned POI tag, associated with each word, to the bidirectional LSTM neural network for training.

### Claim 7 (depends on 1)

The method of claim 1, wherein the POI tags comprise a Begin POI tag, an Inside POI tag, and an Others tag.

### Claim 8 (depends on 7)

The method of claim 7, wherein identifying the POI tags associated with each target word comprises:
assigning the Begin POI tag to a word in the sentence marking a beginning of the POI text;
assigning the Inside POI tag to each word in the POI text succeeding the word marking the beginning of the POI text; and
assigning the Others tag to each remaining word in the sentence.

### Claim 9

A Point of Interest (POI) processing device for extracting POI from natural language sentences, the POI processing device comprising:
a processor; and
a memory communicatively coupled to the processor, wherein the memory stores processor instructions, which, on execution, causes the processor to:
create an input vector comprising a plurality of parameters for each target word in a sentence inputted by a user, wherein the plurality of parameters for each target word comprises a Part of Speech (POS) vector associated with a target word and at least two words preceding the target word, a word embedding for the target word, a word embedding for a head word of the target word in the dependency parse tree of the input sentence, and a dependency label for the target word;
process for each target word, the input vector through a trained bidirectional Long Short Term Memory (LSTM) neural network;
assign POI tags to each target word in the sentence based on processing of associated input vector through the trained bidirectional LSTM neural network;
extract POI text from the sentence based on the POI tags associated with each target word in the sentence; and
provide a response to the sentence inputted by the user based on the POI text extracted from the sentence.

### Claim 10 (depends on 9)

The POI processing device of claim 9, wherein the processor instructions further cause the processor to determine the plurality of parameters for each target word in the sentence inputted by the user.

### Claim 11 (depends on 10)

The POI processing device of claim 10, wherein the response comprises at least one of an answer to a user query and an action corresponding to the user query.

### Claim 12 (depends on 9)

The POI processing device of claim 9, wherein the dependency label for the target word indicates a relation of the target word with the head word in the sentence.

### Claim 13 (depends on 9)

The POI processing device of claim 9, wherein the processor instructions further cause the processor to train a bidirectional LSTM neural network to identify the POI tags for words within sentences.

### Claim 14 (depends on 13)

The POI processing device of claim 13, wherein to train the bidirectional LSTM neural network, the processor instructions further cause the processor to:
assign the POI tags to each word in natural language sentences retrieved from a data repository of natural language sentences comprising a plurality of POI scenarios;
iteratively input assigned POI tag, associated with each word, to the bidirectional LSTM neural network for training.

### Claim 15 (depends on 9)

The POI processing device of claim 9, wherein the POI tags comprise a Begin POI tag, an Inside POI tag, and an Others tag.

### Claim 16 (depends on 15)

The POI processing device of claim 15, wherein identifying the POI tags associated with each target word comprises:
assigning the Begin POI tag to a word in the sentence marking a beginning of the POI text;
assigning the Inside POI tag to each word in the POI text succeeding the word marking the beginning of the POI text; and
assigning the Others tag to each remaining word in the sentence.

### Claim 17

A non-transitory computer-readable storage medium having stored thereon, a set of computer-executable instructions causing a computer comprising one or more processors to perform steps comprising:
creating an input vector comprising a plurality of parameters for each target word in a sentence inputted by a user, wherein the plurality of parameters for each target word comprises a Part of Speech (POS) vector associated with a target word and at least two words preceding the target word, a word embedding for the target word, a word embedding for a head word of the target word in the dependency parse tree of the input sentence, and a dependency label for the target word;
processing for each target word, the input vector through a trained bidirectional Long Short Term Memory (LSTM) neural network;
assigning POI tags to each target word in the sentence based on processing of associated input vector through the trained bidirectional LSTM neural network;
extracting POI text from the sentence based on the POI tags associated with each target word in the sentence; and
providing a response to the sentence inputted by the user based on the POI text extracted from the sentence.
