a keyword analysis tool/API in flask: extracts keywords from e-commerce items in different categories and provides keywords/price distribution information with user’s feedback input;

• keyword-price-api
• Typical Workflow
• API calls
  • [GET] get-keywords
    • Responses
    • Sample Call:
    • Code Workflow
  • [GET] get-pricespread
    • Responses
    • Sample Call:
    • Code Workflow
  • [POST] read-feedback
    • Input Data Params
    • Responses
    • Sample Call:
    • Important Note
    • Code Workflow

Base Path: http://175.106.99.99:16758/

beforehand: upload target database to MongoDB

for i = 1 to repeat do:

{

1. GET-KEYWORDS

   a. i=1: return keyword information for each category in its initial state

   b. i>=2: returns feedback-applied keyword information and most recent feedback file for #3(POST-FEEDBACK)

2. GET-PRICESPREAD: (optional)

   a. i=1: returns a dictionary of item name and img link of histogram depicting item's price distribution for each category

   b. i>=2: after inputting feedback: returns histogram img links with feedback applied

3. POST-FEEDBACK

   a. i=1: posts feedback information based on information from GET-KEYWORDS and GET-PRICESPREAD

   b. i>=2: posts a new feedback by APPENDING new options from the most recent feedback received from GET-KEYWORDS

   Note: Refer to Important Note

}

END IF: retrieved satisfactory information

URL: /api/v1/results/keyword

returns keywords' frequencies and productIds that included the keywords for each category

Successful Response

Code: 200
Content:

• result: array of category-objs

  • category-objects

    attributes	type	desc	uses
    categ	str
    keywords	list of keyword-objects	see below

  • keyword-objects

    attributes	type	desc	uses
    rank	int	rank based on appearance frequencies
    keyword	str	keyword name
    appearance	int	appearance frequencies	*note: app.freq and product_list's value can differ
    product_list	list of ints	productId which had keyword in the title

• previous-feedback: feedback json which was posted most recently ; see read-feedback for relevant data structure

e.g.

```
{
    "result": [
        {
            "categ": "portable_fan",
            "keywords": [
                {
                    "rank": 1,
                    "keyword": "fan",
                    "appearance": 5,
                    "product_list": [
                        "82497194884"
                    ]
                },
                {
                    "rank": 2,
                    "keyword": "paperclip",
                    "appearance": 1,
                    "product_list": [ 
                        "82497194884"
                    ]
                },
        },
        {
            "categ": "dustproof_mask",
            "keywords": [
                {
                    "rank": 1,
                    "keyword": "mask",
                    "appearance": 32,
                    "product_list": [
                        "27224642701",
                        "26883864380",
                        "25295068476",
                    ]
                },
            ]
        }
    "previous-feedback": [
        {
            "categ": "golfbag_set",
            "lprice": 60000,
            "hprice": 150000,
            "sub-cats": [
                "golfpouch",
                "bostonbag"
            ],
            "ignore": [
                "golfb",
                "bKakaofriends",
                "special",
                "basic"
            ],
            "effective": [
                "official_website",
                "high-quality",
                "PLACEHOLDER"
            ]
        }
    ]
}        
```

curl --location --request GET 'http://175.106.99.99:16758/api/v1/results/price-spread'

1. Dataset Prep & Cleaning

   1. Generate the following dataframe obj from target DB in mongoDB
   2. current_cat acts as a pointer showing current item's lowest category level (most specific category level)
      1. i.e) iPhone is the lowest category from [Electronics -> Smartphones -> iPhones]
   3. title column is saved after basic string-cleaning

   i.e. 
   (AMAZON)(NEXT-M)<b>Kakaofriends golf</b> mask 세탁 30회 다회용mask (6395452)
                                   ⬇️
   AMAZON NEXT-M Kakaofriends golf mask 세탁 30회 다회용mask
   

   3. when sub-cats is given in feedback: merges lower category to current category

   i.e. 
   "categ": "golfbag_set",
   "sub-cats": [
       "golfpouch",
       "bostonbag"
   ],
   

   In this case,

   golfbag_set + bostonbag + golfpouch => merged into golfbag_set
   `current_cat` column value is changed to golfbag_set
   

2. Keyword Extraction 1 [when no feedback is given]

   1. save all keywords that appeared in each category's item to dict{Category: Counter(keywords)}

      a. i.e. {Category1 : [('handbags',23), ('luxurious',12), ('pouches',5)]}

3. Keyword Extraction 2 [when feedback is given]

   1. Iterate through all keywords that appeared in each category's item

      a. if price is inside the range [lprice, hprice] -> save keyword to gen_countobj

      b. if price is outside the range => save keyword to sus_countobj

   2. Update Counter object

      a. Remove gen_countobj keywords from sus_countobj

      b. Iterate the remaining keywords again to update ignore, effective.

4. JSON Generation

• generate a dumpable final JSON file based on dict from step 2

  Category A
      {
          "rank": (int),
          "keyword": (str),
          "appearance": (int),
          "product_list": list(int)
      }
  

URL: /api/v1/results/price-spread

returns a dictionary of img link of histogram depicting item's price distribution for each category

Successful Response

Code: 200
Content:

예시

{
    "portable_fan": "https://fkz-web-images.cdn.ntruss.com/price-spread/portable_fan.png",
    "mask": "https://fkz-web-images.cdn.ntruss.com/price-spread/mask.png",
    "laptop": "https://fkz-web-images.cdn.ntruss.com/price-spread/laptop.png",
    "showergown": "https://fkz-web-images.cdn.ntruss.com/price-spread/showergown.png",
    "waterbottle": "https://fkz-web-images.cdn.ntruss.com/price-spread/waterbottle.png",
    "large_umbrella": "https://fkz-web-images.cdn.ntruss.com/price-spread/large_umbrella.png",
    "macbook": "https://fkz-web-images.cdn.ntruss.com/price-spread/macbook.png"
}

curl --location --request GET 'http://175.106.99.99:16758/api/v1/results/price-spread'

for categ in all_categories_as_list:

• generate a fresh DataFrame of rows in which lowest_category==categ in DB
• generate a seaborn.histplot(histogram) from DataFrame's price column
• Upload generated histograms to objective storage using boto3
• save the links as dict{categ: img_link}

end for

return dict

URL: /api/v1/feedback

reads and applies user's feedback for future GET requests

• array of:
  • keyword-feedback object

  • attributes	type	desc	uses
    categ	str	keyword name
    lprice	int	min value in genuine product's price range	item keywords with the price inside genuine's range is removed from final results
    hprice	int	max value in genuine product's price range	"
    sub-cats	listOfStrings	lower category names that will be merged into current category	items in lower category will all belong under current keyword
    ignore	listOfStrings	keywords to ignore	removed from final return
    effective	listOfStrings	keywords that seem promising/critical	displayed even if app.freq is 0

i.e.

[
    {
        "categ": "golfbag_set",
        "lprice": 60000,
        "hprice": 150000,
        "sub-cats": [
            "golfpouch",
            "bostonbag"
        ],
        "ignore": [
            "golfb",
            "bKakaofriends",
            "special",
            "basic"
        ],
        "effective": [
            "official-website",
            "high-quality",
            "PLACEHOLDER"
        ]
    },
    {
        "categ": "hats",
        "lprice": 35000,
        "hprice": 55000,
        "sub-cats": [],    # they can be left as empty list
        "ignore": [],
        "effective": []
    }
]

Successful Response

Code: 200
Content: received feedback

curl --location --request POST 'http://175.106.99.99:16758/api/v1/feedback' \
--data-raw '[
    {
        "categ": "hats",
        "lprice": 35000,
        "hprice": 55000,
        "sub-cats": [],
        "ignore": [],
        "effective": []
    },
]'

For any iteration(i>=2) mentioned inTypical Code Workflow, you must include previous-feedback from [GET] get-keywords so all the previous options can be applied for the new iteration.

Therefore when writing new feedback files, we suggest you copy previous-feedback's content and edit accordingly.

i.e.

"previous-feedback": 
[
    {
        "categ": "golfbag_set",
        "lprice": 60000,
        "hprice": 150000,
        "sub-cats": [
            "golfpouch",
            "bostonbag"
        ],
        "ignore": [
            "golfb",
            "bKakaofriends",
            "special",
            "basic"
        ],
        "effective": [
            "official-website",
            "high-quality",
        ]
    }
]

updated feedback.json

[
    {
        "categ": "golfbag_set",
        "lprice": 60000,
        "hprice": 150000,
        "sub-cats": [],             # editted
        "ignore": [
            "golfb",
            "bKakaofriends",
            "special",
            "basic",
            "discount",                 # added
            "golf"                  # added
        ],
        "effective": [
            "official-website",
            "high-quality",
            "made_in_china",               # added
            "imitation"                   # added
        ]
    }
]

• "sub-cats": remove options as lower categories were merged in the previous iteration
• "ignore": added 'discount' and 'golf'
• "effective": added "made_in_china" and "imitation"

- read the POSTed json file
- save as `./cache/feedback.json` & `./cache/feedback.pkl` for future accesses