Turn Commenters To Buyers For Ecommerce Brands

• 

  Results of different tests

Inspiration

For years the conventional way to advertise on social media platforms(fb instagram etc..) for e-commerce brands are:

1)Run an advertisement 2)See who are clicking the ad and buying 3)Click ad not bought, retarget them etc.. ..

From this data tiktok algorithm like any other social networks then suggest e-commerce brand owners with similar products to advertise on similar/same type of people who were interested.

BUT

Tiktok has something very powerful other social networks don't have.

Organic videos which are essentially not an 'advertisement' such as these hashtags(unboxing amazonfinds,amazonmusthaves,aliexpressfinds,petsmusthavesamazon) which are essentially most powerful way of advert.Organic.

I wanted to find a way for tiktok to leverage this huge opportunity using comment section essentially of these type of videos.

This would help tiktok to understand what type of organic videos/creators/products leads to a high buying intent overall.

Internal Usecase:

Identifying a product with high buying intent in comment section lets tiktok algorithm to take into consideration when a new ecom brand with similar product comes to target people on comment section.

External Usecase:

This data publicly available could be helpful for letting ecommerce brands to partner with these not influencers but creators who creates a demand and buying intent instead of just picking 'influencers' their following/engagement rate, a new metric introduced BUYING INTENT.

What it does

It leverages large language models to understand buying intent from a sentence given in comment section.

How we built it

I have gathered data from comment sections of: unboxing amazonfinds,amazonmusthaves,aliexpressfinds,petsmusthavesamazon

Manually tagged what is a high buying intent based on a sentece and what is not to train my model with this data.

Then pre-trained a bert model to understand what corresponds to a high buying intent and what not.

Challenges we ran into

1)Data which i wanted was not available to me since it requires research api access to use tiktok api and gather more data from comment sections.

In result my model was trained based on a limited amount of data.

2)When labelling data based on high buying/low buying intent was only labels i could have used to train model.I tried having a numeric 1-10 representation of training my dataset based on intent level.

Only 0 and 1 was available to train,which resulted poored decision making for my model.

Accomplishments that we're proud of

Based on the input data provided which is not as large to train with a higher accuracy..but our model works to predicting comments mostly.

What we learned

DATA.Better input data source better prediction algorithm.

Cleaning up data better results in much better accuracy.

Removing unwanted data results in better training,better accuracy.

Diversity of comments in training data set results in broader understanding to model and better results.

What's next for Turn Commenters To Buyers For Ecommerce Brands

Tiktok giving me access to research api so i can have access to download all the comments available related to the hashtags and improve model accuracy by improving my training dataset.

Built With

• apify
• bert
• largelanguagemodel
• pandas