Conscious Recommendation System (CRS) is an intellectual property in the data science sphere aimed to optimize and improve the internet-trading profitability.
The system can be used in the existing online stores as well as in the newly created ones.
With the help of data mining, the CRS provides the end user with precisely targeted and conscious recommendations for those goods that seem really appealing to the customer during the buying process. A conscious recommendation represents the final result of data analysis system about consumer decision-making process during a long-term period.
Areas of application
Functional characteristics
General operation principles – CRS
System use in business
What kind of information does the CRS need?
How to connect your store to CRS?
The recommendation system is used in the retail sphere where there are stable combinations of the sold goods (in this case the system is capable of producing the conscious recommendations after learning).
The system was originally developed for the pharmaceutical retail sphere since the stable combinations of medicines and medical supplies are widely used for the treatment of various diseases there.
The system can be also used in the following spheres:
The main difference between the CRS and the traditional recommendation systems is that the CRS makes recommendations for creating a complete list of the in-demand set of goods at a certain moment. The recommendations of type “the best-seller”, “people also buy” or “you used to buy this” are taken into account with different weighting factors while showing the results to the clients.
For example, if a customer has chosen some alpine skis, the system will select and recommend the most suitable goods related to this particular product – a pair of ski boots, some gloves, a pair of ski poles etc. And, as well as in a case with medicines (recipes), the recommendations are based not on the frequency of goods appearance but relying on a research of all consumer demands for similar sets of goods. Such a behavior of the system is grounded on the usage of NLP (Natural Language Processing) principles in regards to the demand research.
The general recommendations are also taken into account by the system, and, basing on the wishes of the commercial network management are added to “conscious recommendation” increasing the marginality and specific goods promotion.
Within the system, we have also found the solution for so-called “cold start” (is used in case there is not enough information to offer for a certain choice). The majority of recommendation systems has some problems when providing the results to the new customers or when specifying the related goods for new items. We have developed the new methodologies for covering these needs.
Providing the commercial network management with the list of the products that could have been bought by the buyers but were out of stock (“drug shortage” in medicine) is another important and fundamental difference between our recommendation system and the analogs.
Implementation of the above-mentioned functionality is possible because the system predicts the items that can be added to the cart and checks their availability. The shortage list gets constantly updated before the user gets any recommendation on the website when checking the availability of the product in stock and getting it replaced if needed.
CRS accommodates the client’s needs and predicts the largest number of goods that can be purchased along with the first product that was selected. Once new items are added to the cart, the system generates the recommendations based on the entire list of the goods, not on the first or the last ones only.
Fig.1. The main functional blocks of CRS
At the same time, the system tries to find such goods that belong to the largest possible subgroup from the cart. For example, if a pair of skis, ski bindings, a pair of gloves and a soccer ball are in the cart the system will detect that the first three goods are the main ones and will recommend buying a pair of boots, ski poles, and other possible options. At the same time, the recommendation “ball pump” will also get generated but will not probably be shown to the client (since the purchase would be categorized as skiing group). If the users will not be interested in one of the suggested items from skiing group and will skip it (e.g. nobody wants to buy the recommended ski boots), the system will change it to the “ball pump” in the recommendation. The results will be monitored and if the sales of pumps get increased, the system will take it into account and reorganize the recommendations to optimize the trading process.
The main functional system block that predicts the shopping cart, in general, helps to avoid one of the most important risks – so-called “cold start” (the situation when a merchant knows nothing about his client after the first purchase and has nothing to offer except “related to this item” goods or an equivalent of the recently chosen item). In this case, the CRS provides the list of conscious recommendations in which the first selected product remains the main one. After every other item is added to the cart, the system recommendations get even more “conscious” and appropriate.
Every functional block considers different trade chain needs while organizing the workflow:
If the client is registered in the system and has ta certain shopping history, the system will consider the following factors:
During the consultation with a buyer the seller receives a cumulative list of the goods recommended for sales, based on the first named product. If seller has initialized the buyer in the system, the list is supplemented with a block of user preferences.
Two adder units which combine different issuances enable adjusting each specific version of the system by changing the weighting factors. System is adjusted to the priorities of the trading network, namely the promotion of certain products, user preferences admission, and pre-sales volumes increasing.
Getting these priorities and factors, the system is being trained in process, thus improving the quality of recommendations for a particular business by analyzing the value of recommendations. Different templates for finding useful recommendations compete with each other, and the preference is given to those templates that are most interesting to users and those that generate the larger flow of additional sales.
CRS needs information about current sales, by which it learns the changing market situation. In addition, CRS must get the information about the absence in the warehouse of the goods, identified as recommended by the system. This allows forming the lists of goods that you want to have in stock so to correctly plan the purchases.