Similarity value differs for api call and api demo on UI

[Sergei Zagriychuk]

I need to get similarity score for 2 texts.

For example:

text1 = Don't miss the healthy recipes, suggested workouts and reviews of "We tried it!

text2 = Don't miss out on healthy recipes, recommended workouts, and 'we tried it!' reviews.

When I use api demo via UI https://dandelion.eu/semantic-text/text-similarity-demo/ I'm getting 100% of semantic similarity

At the same time when I hit GET api endpoint it returns me only 0.7435

Here is the call:

https://api.dandelion.eu/datatxt/sim/v1?token=*************************&text1=Don%27t%2520miss%2520the%2520healthy%2520recipes%2C%2520suggested%2520workouts%2520and%2520reviews%2520of%2520%22We%2520tried%2520it%21&text2=Don%27t%2520miss%2520out%2520on%2520healthy%2520recipes%2C%2520recommended%2520workouts%2C%2520and%2520%27we%2520tried%2520it%21%27%2520reviews.&lang=en

Response:

{"time":2,"similarity":0.7435,"lang":"en","timestamp":"2019-05-03T14:35:29.167"}

How is it possible? Am I missing something?

[ber...@dandelion.eu]

Hi Sergei,

the demo should perform a "plain" call to the similarity API.

Please:

- check if you are setting the `bow` parameter in your calls

- try to make POST http calls instead of using GET method (check how the url is escaped)

- in general check if some pre-processing step modifies your text without "your control" (e.g., double quotes are removed)

On your example, performing a POST call, I obtain the result of the demo.

Here the code you can execute form a shell:

curl --request POST \

  --url https://api.dandelion.eu/datatxt/sim/v1 \

  --header 'content-type: application/x-www-form-urlencoded' \

  --data 'token=***************&text1=Don'\''t%20miss%20the%20healthy%20recipes%2C%20suggested%20workouts%20and%20reviews%20of%20%22We%20tried%20it!&text2=Don'\''t%20miss%20out%20on%20healthy%20recipes%2C%20recommended%20workouts%2C%20and%20'\''we%20tried%20it!'\''%20reviews.'

Let us know if testing with the hints above helps you somehow

Regards

Giacomo Berardi

Dandelion team

[Sergei Zagriychuk]

Hello Giacomo,

Many thanks, using of POST instead of GET helped me out!

Thank you,

Sergei