Product Related FAQs 💡

How does the review summarization work?

Review summaries is an AI service offered to Bazaarvoice clients, which provides product summaries using consumer reviews. The summaries are generated natural language processing (NLP) techniques and large language models (LLMs), to generate concise summaries of customer reviews. Instead of reading through numerous individual reviews of a given product, users can read a short summary that captures key themes, opinions, and sentiments from a larger set of reviews.

Determine eligibility for use?

It is available only for clients on Advanced and Enterprise packages having Display in their package.

Which product qualifies for a review summary?

The minimum number of reviews required to generate a summary is 30 reviews in English language. This is to ensure a breadth of consumer feedback has been provided, when generating the review summary.

Why might a product with 30 reviews not have a summary?

While a product may display 30 or more reviews, the review summary generation requires a minimum of 5 approved English language reviews from the native product or one of its source products. If a significant portion of the displayed reviews are syndicated, the native review count might not meet this threshold. This requirement ensures the quality and relevance of our summaries.

How often are review summaries updated?

Review summaries are updated when the number of moderated English reviews increases by 20% (rounded to the nearest integer) from the last summary generation. For example, after the initial 30 reviews, the next update would occur at 36 reviews, and so on. This threshold ensures that summaries are only regenerated when there's significant new consumer feedback. The update process for summaries based on this logic will be run at least every week.

Are syndicated reviews included in the review summaries?

Yes, review summaries incorporate native, family, and syndicated reviews that have been approved through our moderation process. This comprehensive approach ensures that summaries reflect the full range of consumer experiences across different platforms.

What is the expected summary length?

The summaries are up to 100 words in length. There are two formats available to chose from: a paragraph formatted summary; and a bullet point formatted summary. Please refer to the API reference for more details.

Does review summarization preserve the sentiment of the original reviews?

Yes, the review summarization model has been evaluated against relevancy and representativeness against the original product reviews. The summarization process aims to capture the overall tone of the reviews, such as positive, negative, or neutral feedback. As summaries are concise representations of many reviews, not every topic in every review can be covered, particularly for products with very large volumes of reviews. However, we evaluate our process using large evaluations sets and several evaluation metrics to ensure the summary reflects the range of opinions of the product’s consumers.

Who owns the output from the review summarization process?

The generated output (review summary) is a service provided by Bazaarvoice. Ownership of the output will reside with Bazaarvoice, with the service being discontinued at the end of the client contract.

Can review summarization be used for multiple products or services?

Review summaries are generated on an individual product basis, so yes summaries are available for multiple products as unique summaries. Clients are welcome to analyse or compare and contrast product review summaries, across products, brands or even categories and sub-categories.

Can negative reviews be eliminated while generating the summaries?

Bazaarvoice's first principle is Authenticity and Trust. Therefore, the model takes into account all reviews of a product, including positive, negative, and neutral ones. The summaries generated are based on all reviews and are completely authentic.

Why might there be differences between displayed reviews and the summary?

There can be timing differences between when reviews are displayed and when summaries are generated. Reviews may be added, edited, or removed after a summary has been generated. The extent of these differences may vary depending on how clients implement reviews and summaries in their systems. Our update logic helps manage these timing differences.

AI Model, Data Security, Privacy and Governance 🕸️

How safe is your data?

Input data is hosted and stored in a secure AWS environment. We prioritize data security and privacy. All processing is done within our secure infrastructure, and we do not share client data with third-party providers outside of our secure cloud platform.

Model limitations and hallucinations?

Currently, the model only supports the English Language. Additional languages will be considered for future releases.

We evaluate the text summarization process with a suite of automated and human-in-the-loop, both offline and online to ensure hallucinations are avoided.

What privacy mechanisms are in place to ensure PII used in connection with the AI System adheres to data privacy/protection laws?

Review summaries are solely based on moderated reviews, following Bazaarvoice's extensive automated and human-moderation process. This ensures the summaries are based on approved reviews free of personal information or other non-acceptable content. These mechanisms ensure the aggregate summaries adhere to Bazaarvoice’s data privacy and protection policy.

Can the model accuracy improve over time? Can the models be retrained post accuracy testing on live data?

Our Review Summaries feature is built on a robust and evolving AI framework. We continuously monitor the performance and quality of our summaries to ensure they meet high standards of accuracy and relevance. Our team of AI experts and machine learning engineers regularly evaluate opportunities to enhance our summarization process, which may include refining our algorithms, updating or fine-tuning our language models, or incorporating new AI technologies as they become available.

While we're always working to improve our service, any updates or enhancements are implemented systematically across our platform to maintain consistency and quality for all our clients. We do not currently offer customized model training or client-specific alterations to our summarization process.