Harnessing NLP for Real-Time Contextual Advertising Solutions
Written on
In the contemporary landscape of online advertising, efficient marketing strategies have traditionally depended on various user tracking methods, including third-party cookies and device fingerprinting. However, with increasing concerns about data privacy and the introduction of regulations like GDPR and CCPA, these methods are becoming outdated. Browsers like Safari and Firefox have implemented features to limit cross-site tracking, and Chrome is also developing alternatives. Consequently, the era of third-party cookies is nearing its end. Apple's Identifier for Advertisers (IDFA) will soon require explicit user consent for app usage. As cross-domain tracking fades, advertisers are turning back to contextual advertising.
This article outlines how to utilize the Text Classification API from NLP Cloud for implementing contextual targeting. The approach can be adapted for various advertising technologies, including ad servers and OpenRTB.
Contextual Advertising
With the decline of third-party cookies, the resurgence of contextual advertising campaigns is a logical prediction. This method allows advertisers to target user interests at scale by focusing on the content users are currently viewing, rather than relying on their browsing history or behavioral profiles.
This strategy is likely to be more engaging for users, as they will encounter ads relevant to the topics on the websites they visit.
The Importance of Tags
Most advertising technologies and networks facilitate the use of keywords or tags during ad serving. Text remains the foundational element of the web and is a valuable source of information. However, extracting context, tags, and keywords for advertising or recommendation purposes can be challenging and labor-intensive. For owners of medium-sized news websites, extracting relevant topics beyond a few editorially assigned tags can be quite difficult.
Early attempts to automate this process often led to amusing errors:
Advancements in Natural Language Processing (NLP) now enable more precise and efficient text classification. This involves categorizing text content into relevant labels or categories.
Consider a webpage featuring articles on diverse subjects:
The goal is to ensure that advertisements displayed are thematically linked to the article content.
Criteria for Our Solution: 1. Select keywords and topics relevant to the advertising campaign. 2. Analyze the content of the visible article and categorize it accordingly. 3. Choose and display relevant advertisements.
Note that while advertising systems and web development are not the focus of this article, the underlying concepts apply across various tools and technologies.
Text Classification API
I recommend using a dedicated API for managing text classification logic. We have two choices: develop our own or utilize an existing solution.
Creating a simple text classification engine with Python and NLP libraries can be accomplished in an afternoon. However, challenges arise regarding accuracy and the ability to handle increased traffic. It's essential to manage the growing user base and their interactions effectively.
Website owners typically prefer not to engage in fine-tuning machine learning models. Therefore, it's practical to delegate as much as possible to an external service. Importantly, we do not plan to transmit any user data, only information related to the website, simplifying privacy concerns when using external contextual targeting tools.
NLP Cloud offers various APIs for text processing utilizing machine learning models, including a text classification API that appears user-friendly for straightforward implementation.
With the NLP Cloud API, you can test which algorithm suits your specific business needs.
Integrating Text Classification with Website Content
Given that the backend of our website operates on Python (Flask), we initiate the process by developing a straightforward client for the NLP API:
Results:
The outcome is promising, with each label being accurately assigned its relevance to the topic effortlessly.
The plan is for the ad serving system to manage the selection of displayed banners based on the scores assigned to the relevant labels. To protect API keys and maintain control over data, we will implement a simple proxy.
Campaigns
Assuming we have three advertising campaigns to execute:
Next, we will outline a mechanism on the frontend to manage the display of the appropriate advertisements.
This is our ad server?
Utilizing fetch, we will retrieve labels for the article text based on its selector:
Note that we will only show client ads if the score exceeds 0.8:
Math.max(...scores) >= 0.8
If not, we will display self-promotional content.
This threshold can be adjusted based on specific needs.
The attentive reader might notice that the example for the hairdresser's advertisement did not appear. This is due to the serious nature of the articles, which do not align with fashion-related topics. To implement this campaign, a different site or keyword strategy may be necessary.
Performance Considerations
One significant advantage of using asynchronous fetch is the improved loading speed of pages. However, ads will only appear after the labels are downloaded. To mitigate costs and enhance efficiency, implementing a caching mechanism in a production environment is advisable.
Another option could involve storing labels directly in the database, particularly for articles that do not change frequently.
Using a dedicated API to assign labels to any text allows us to employ JavaScript on virtually any page in near real-time, even without backend access!
Key Takeaways
The primary challenge with contextual targeting on news websites lies in the diverse topics presented in the articles, including those relevant to advertisers. However, the often sensational and serious tone of news can be unsuitable for advertising.
The Text Classification API from NLP Cloud effectively tags texts, enabling us to repeat the process while excluding specific topics from ad placements.
If you found this article insightful, please consider supporting the charitable foundation I'm involved with: https://4fund.com/w7ctjx