AI can de-clutter crowd wisdom in online marketplaces

Buying online is getting complicated. Wasn’t it supposed to simplify? The pendulum has swung the other way. Why? There is just too much data and it’s hard to decipher.

Can you give me an example? Take one of the top global mobile brands and its presence on one of the most valued eCommerce sites in the world. You will find 216 models which have 4,013,852 reviews and 667,868 ratings. And this is one site. Wow!! 4 million reviews… it’s a non-starter… which human will read it to purchase a phone?


It is going to get worse. 47 Zeta Bytes of information is expected to be generated by 2020. Only 16% of this is expected to be structured & searchable. A lot, of course, will be fake information that we are struggling with today.

We need to De-Clutter and do it intelligently to leverage the crowd wisdom behind this data.


Yes, let’s see what is possible and then a bit on how. The how is also explained in my previous blog along with more examples on de-cluttering life (AI to de-clutter life).

Note: to keep the focus of discussion on the concept & AI, all brand names are covered/ blacked out

Can we use the ratings we find on eCommerce sites?

Yes and no. The pendulum has swung in this too, increasingly, across product categories. Going back to the mobile example above let’s write a piece of code to analyze the ratings and reviews. See Figure 1 below.

Figure 1 (All charts are auto-generated by code & AI)

Key takeaways:

  1. 60% of products have 4.1–4.4 rating on a scale of 5. That’s useless, when ratings crowd up it provides no meaning.

  2. As people rate a product more, there is an increasing chance the product will have an average rating.

  3. High-value products have fewer reviews. This is exactly where one needs more reviews as the spend is more but as there are fewer buyers - there are fewer comments too.

Crowding of ratings and lack of reviews for higher value products just adds to clutter rather than reducing it. It is increasingly a distraction to consumers.

Let me provide one more data point to show that ratings are increasingly a distraction. Let’s take the simplest of things. Deciding to go to a restaurant. Most restaurants these days have 3000+ reviews and 1000s of Instagram images. Let us use one of the biggest restaurant review sites out there, Yelp, to check ratings. Note this: 48% of Yelp reviews are 5 stars and those can’t be differentiated. 68% of reviews have 4+ stars (Source: And 72% will have a recommended rating. Clutter. Ratings are increasingly a distraction and clutter. Nothing that helps me decide unless I wade through the comments.

That leaves you with crunching comments/ reviews at scale.

What if you got a dashboard like below (Figure 2) where AI has understood all the comments about what to eat, see and expect?

Figure 2 (All charts are auto-generated by code & AI)

It’s a harbor facing restaurant in San Francisco with views of Alcatraz and the Pier. Clam Chowder, crab cakes, and mixed grill salmon shrimp are must-have items on the menu. Slightly touristy. Great service and subtle aspects like in particular look for waiter service from Tyler. 4000 comments and 56 menu items summarized for your convenience. De-Clutter.

Is this possible?

Yes, and repeatable. We can have an AI build a language model that understands comments and identifies various topics/ conversations in those comments and bring forth the most important ones.

Can we do for mobiles?

Yes, let us go back to the eCommerce discussion and look at mobiles. We had via coding found that 60% of phones had 4.1–4.4 rating. Roughly 40% we had thus eliminated. Let us focus on this 60% and crunch comments and identify phones that have been returned after ordering. AI provides the below figure (Figure 3)

Figure 3 (All charts are auto-generated by code & AI)

The brand seems to have heating issues and is a major cause for product returns. Some issues in the covering case of the mobile and value for money perception. This can help us eliminate all mobiles with such bad reviews. Giving us a much smaller subset to work on. By playing with input data and using the same AI platform one can take multiple cuts at coming to a decision.

If I finally shortlist a product can I get a restaurant type dashboard?

Here you go for a particular mobile. Figure 4 is an AI provided dashboard.

Figure 4 (All charts are auto-generated by code & AI)

Exactly similar dashboard as the restaurant. While the positives can be highlighted as in the restaurant example, here, a different aspect is illustrated which is the “issues needing attention”. The conversations/ topics can be obtained to understand the product/ service like earlier but further crunched to find the sentiment around each. This mobile while good seems to have touchscreen issues including clarity. There you go we zoomed into what is the most important negative aspect of this phone. Now you decide whether to buy or not. Rest is clutter.

What about the various features? Can AI address those?

Let us look at one of the most sold electronic gadgets in the world today. Figure 5 shows a similar dashboard but here we do a feature wise analysis. The product sells and one can see why… across features, people have a very positive sentiment in addition to a very positive sentiment across topic categories. The comments can be used to provide metrics like Net Promoter Score (the difference between the percentage of Promoters and Detractors) which can give the consumer a simple one look view of product acceptance.

Figure 5 (All charts are auto-generated by code & AI)

Can we see more varied product categories?

Below is a washing machine. Figure 6 suggests that this is a good washing machine. While delivery is good there seem to be teething issues in installation. Especially people who come for installation tend to ask for additional money and also try to sell accessories like a stand to keep the machine. Just zoom into what matters instead of the 12148 ratings and 2502 comments. De-clutter.

Figure 6 (All charts are auto-generated by code & AI)

Oh, this is still electronics what about other categories?

Let us crunch comments for an ethnic dress like sari which is worn in India. The AI dashboard (Figure 7) says that this sari which has a pink border actually arrives as a red bordered sari. It also has fabric issues. As per the crowd wisdom, the sari still looks beautiful so if the quirks are OK then go for it.

Figure 7 (All charts are auto-generated by code & AI)

How is all this accomplished?

The De-Clutter Life platform can take varied forms of inputs and crunch it using various AI techniques to figure out conversation/topic flows and pull out important conversations as de-cluttered output. The AI black box is a combination of unsupervised learning language models and flow models, deep learning based language models, classifiers and named-entity recognition neural networks topped with heuristic algorithms to extract dashboards. Extractive summaries of issues etc are obtained by running page-rank kind of search algorithms on the language models. Additional details are in my previous blog along with more examples on de-cluttering life (AI to de-clutter life)

Figure 8

As data explodes de-clutter of online commerce platforms like eCommerce, online food delivery portals, travel websites, job sites etc is possible while maintaining the power of crowd wisdom. AI can de-clutter.

Data holds the truth. Code+AI can unlock it. In the world of AI… data matters, intelligence matters.

Note: Companies interested in implementing the above IP in their organization/ offerings can contact via Linkedin

6 views0 comments

Recent Posts

See All