Soft Bing advertising recently launched a new method to expand the exposure range of keyword advertising, which can automatically include the relevant search words not on the keyword advertising list into the advertising exposure range, and even the keywords not under the advertisers can also be exposed. Microsoft currently puts this technology to a fashion retailer for trial. Microsoft claims that it can effectively improve the exposure range after the experimental results are confirmed The high click through rate was 5.9%, but Microsoft did not disclose the details of the experiment or the company name of the experimenter.
In addition, this technology can not only be used in the search results of advertising, but also in the internal search of the website, through the combination of personal buying behavior or search history, customized shoppers’ personal taste search, to improve the relevance of the overall search results.
In traditional search ads, advertisers specify keywords, and the system will match queries and product descriptions to rank related product ads, but this way is not feasible in the fashion industry. Microsoft mentioned that if advertisers want to specify and bid for all possible fashion keywords, the cost will be very expensive, and the selection of keywords also determines the product advertising Ranking methods and keywords affect the operation of the mechanism to maximize relevance, thus limiting the number of times users click these ads.
The keywords of users in search engines may be < brand name > dreams on sale, while the keywords of advertisers are <brand name> and dreams, and the descriptions of these products are <brand name> bodycon MIDI dreams and <brand name> jacquard MIDI dreams, etc. in traditional methods, advertisements only rank by matching search queries with advertising keywords, which will lose bodycon MIDI and jacquard MIDI are valuable information, but the results are not accurate enough.
Microsoft’s solution to this problem is to go the other way, so that the search ranking for fashion retail is not affected by keyword matching, but based on unmatched words of product description that do not meet the query keywords, ranking the ads. In the product description, the importance of unmatched words can be learned from the same queries in the past and users’ clicks on other products. Similarly, if the click rate of < brand name > jacquard MIDI dress is higher than that of < brand name > bodycon MIDI dress, it can be concluded that the importance of the unmatched product description word jacquard is higher than that of bodycon. In some cases, the importance can be inferred as the degree of popularity, and we can get the conclusion that jacquard is more popular than bodycon.
In order to facilitate similar queries, Microsoft will bind the search keyword dresses with unmatched product description words to become query product word pairs, and generate data such as dresses jacquard and dresses bodycon to train data training machine learning model, so as to learn the mode of query product and click through rate.
When the user enters a query that has never been seen before, the model will generate the corresponding query product words for the query and product description, and be used to predict the click rate of each product. The products with the highest click rate are predicted to get the highest ranking, and the advertisements of products can be arranged in descending order according to the predicted click rate to maximize the relevance of advertisements and the click times of query users.
Adding unmatched words to the ranking consideration not only avoids the trouble that fashion industry advertisers have to manually specify all keywords, but also helps fashion retailers evaluate the importance of query advertising keywords originally used for bidding.
At the same time, this technology can not only be used in the basic advertising ranking of search engines, but also in the search of fashion retailers’ websites. Microsoft said that because of their subjective taste, retailers can use the technology to tailor search recommendations for each customer, or even add historical purchase lists as reference.
This technology can meet the unique consumer behavior of fashion retail, and provide better prediction based on the products clicked by users in the past, combined with seasonal fashion trends, or even constantly changing aesthetics.