Billions of dollars are spent on search engine optimization and search engine marketing every year. A new study published in the INFORMS journal Marketing Science explores the complications of the strategy, shining a light on the inherent challenges that come along with using search terms with multiple meanings.
Authored by researchers from the Hong Kong University of Science and Technology and Columbia Business School, the study focuses on understanding what consumers mean, or intend, when they type in specific search terms.
Consumers frequently use the same terms in different ways, making it a challenge for marketers to accurately understand their online queries. Professors Jia Liu and Olivier Toubia found that digital marketers are at a distinct disadvantage as they attempt to infer content in quantifiable ways. Liu and Toubia found that inferring content preferences from search queries presents challenges first and foremost because search terms tend to be ambiguous by nature.
“With 3.5 billion searches on Google per day, search advertising now is tremendously important for marketers. Its success hinges on marketers’ ability to infer the search context underlying short search queries, and then design advertising campaigns accordingly,” Toubia says. “This motivated us to develop a method to extract ‘topics’ from search queries and search results, which could provide that context.”
Three main challenges faced by marketers were highlighted in the study. First, consumers use the same search terms in different ways. Second, the number of possible keywords that consumers can use is incredibly high. Third, most search queries are very short, containing up to just five words.
So what’s the solution? Liu and Toubia identified a new approach to provide better content for individual search terms. Using a “topic model,” powered by a learning algorithm that extracts topics from text based on the occurrence of the text, marketers can quantify the mapping between queries and results. Liu and Toubia’s model is designed to create context when one term is semantically related to another.
Comparing existing topic models with their proposed topic model in the application, the researchers found that their model was better at inferring a consumer’s true intentions and preferences, based on observing a single search query by that consumer.
“This is because our model helps combine information from multiple search queries and their associated search results and quantifies the mapping between queries and results,” Liu says.
Toubia says he was pleasantly surprised to see that the model he and Liu established was able to improve click-through rate prediction, based on measuring the fit between the copy of the search ad and the consumer preference revealed by their search queries.
“This improvement is above and beyond the predictions made on the basis of the position of the ad, its quality score, and several other factors,” Toubia says.
The model may eventually be used to explain and predict consumer click-through rates in search advertising, based on the degree of alignment between search ad copy and search engine results pages. The ultimate goal here would be to create a way for marketers to better match actual search results with what users mean or intend when specific search terms are entered into Google.
“The few words in a search query provide a window into what consumers are truly searching for,” Toubia says. “With the right analytics tool, search queries can be mapped onto a complex set of consumer interests and needs. This can help design more effective search engine marketing and search engine optimization campaigns.”
Stephanie Miles is a senior editor at Street Fight.