[PYTHON] Make a chatbot and practice to be popular.

Introduction

In recent years, most of communication is shifting to chat such as LINE. In today's society, where it is necessary to write attractive sentences in order to be popular, there are limited places to practice due to the surrounding environment, and many people will suddenly come to the actual performance. ~~ I am. ~~ To address the challenges of humankind living in the present age, we will prepare a chatbot that outputs the degree of interest in ourselves, practice throwing messages to it, and take an approach to improve our message skills.

The figure below is a part of the training that the author is conducting.

The degree of interest of the other party is output in three stages (positive, uninterested, unknown). The other party's message itself has no meaning. I prepared what I have been told.

You win if you make your opponent positive. It seems that the actual training will not come forever even if you practice here, but in this article, we will explain how to create a chatbot and aim to improve the message skills of humankind.

[Updated on February 23, 2020] I made a LINE bot.

create

Even if I try to create it, the opposite sex is not familiar in the first place. In other words, the degree of interest cannot be measured. Meanwhile, the author discovered married / unmarried ones in the user attribute estimation of the reference of COTOHA API, which is a natural language processing platform. For those who are not popular, married people have performed miracles that cannot normally be accomplished, that is, God. I thought. According to the estimation by machine learning, it is estimated that the sentence is married, that is, it is a message that attracts interest. Natural Language Processing / Speech Processing API Platform COTOHA API

Using the COTOHA API like this, create a bot that outputs (positive, uninterested, unknown) to your message.

Throw to COTOHA API

First, register with Developers from the COTOHA API link above. The free tier is 1,000 calls / day each. (As of February 13, 2020) With all this, you can probably practice without rest.

The minimum requirement is

1. Obtaining an access token
2. Throw the entered string to user attribute estimation

This time, as a room for customization, if the estimation result is "unmarried" or the result is not returned, a message that you have seen is returned at random. In addition, when "married" is encountered, the input character string is thrown to "sentence type estimation", and a relatively favorable message is returned according to the type of speech act. This will make the story seem connected and the painful discipline will be a little lighter.

import random
import requests
import json

BASE_URL = "https://api.ce-cotoha.com/api/dev/nlp/"
CLIENT_ID = "******"
CLIENT_SECRET = "******"

def auth():
    token_url = "https://api.ce-cotoha.com/v1/oauth/accesstokens"
    headers = {
        "Content-Type": "application/json",
        "charset": "UTF-8"
    }

    data = {
        "grantType": "client_credentials",
        "clientId": CLIENT_ID,
        "clientSecret": CLIENT_SECRET
    }
    r = requests.post(token_url,
                      headers=headers,
                      data=json.dumps(data))
    res = r.json()

    return res["access_token"]  # ["expires_in"]:remaining time

def sequence_anlysis(sentence):
    headers = {
        "Content-Type": "application/json",
        "charset": "UTF-8",
        "Authorization": "Bearer {}".format(access_token)
    }
    data = {
        "sentence": sentence,
        "type": "default"
    }
    r = requests.post(BASE_URL + "v1/sentence_type",
                      headers=headers,
                      data=json.dumps(data))
    res_type = r.json()
    if not res_type["status"] == 0:
        print('type API Error status', res_type["status"])
        exit(1)

    data = {
        "document": sentence,
        "type": "kuzure",  #Collapse mode such as tweets
        "do_segment": False
    }
    r = requests.post(BASE_URL + "beta/user_attribute",
                        headers=headers,
                        data=json.dumps(data))

    res_attr = r.json()
    if not res_attr["status"] == 0:
        print('attr API Error status', res_attr["status"])
        exit(1)
    
    if not "civilstatus" in res_attr["result"].keys():
        civilstatus = "unknown"
    else:
        civilstatus = res_attr["result"]["civilstatus"]
    return res_type["result"]["modality"], res_type["result"]["dialog_act"], civilstatus

Prepare a chatbot

Use Python to prepare a chat space that runs on the console. ~~ It was troublesome to prepare LINE Bot etc., so this time I implemented it with standard input and output.

print('End with quit')
while True:
    s = input('you: ')
    if s == "":
        continue
    elif s == 'quit':
        break
        exit(0)
    else:
        #Parse input
        modality, dialog_act, civilstatus = sequence_anlysis(s)
        #Create a reply
        output = make_sentence(dialog_act, civilstatus)
        if civilstatus == "married":
            print('bot (Positive):    ', output)
        elif civilstatus == "Unmarried":
            print('bot (Not interested):  ', output)
        else:
            print('bot (unknown):      ', output)

Summary

A chatbot using the COTOHA API was used to create a training ground for popularity. I sincerely hope that this will save some of the love refugees around the world.

Future tasks

• This time, the degree of interest is output independently for each message. I want to support time series in the future.
• ~~ This time, we are implementing with standard input and output, but if you are an enthusiastic practitioner who wants to practice with LINE Bot etc., please make it public. ~~
• Responses are prepared by unpopular male students from their own discretion and a small amount of past experience. By all means, I would like a chat expert to improve it.