Conditional Semantic Textual Similarity (C-STS) measures the semantic proximity between text segments under a specific condition, thereby overcoming the ambiguity inherent in traditional STS. However, existing methods are largely confined to discriminative models, failing to fully integrate recent breakthroughs in the NLP community concerning Large Language Models (LLMs) and Reinforcement Learning (RL). RL is a particularly well-suited paradigm for this task, as it can directly optimize the non-differentiable Spearman ranking metric and guide the reasoning process required by C-STS. However, we find that naively applying listwise RL fails to produce meaningful improvements, as the model is overwhelmed by complex, coarse-grained reward signals. To address this challenge, we introduce PoLi-RL, a novel Point-to-List Reinforcement Learning framework. PoLi-RL employs a two-stage curriculum: it first trains the model with simple pointwise rewards to establish fundamental scoring capabilities, then transitions to a hybrid reward that combines pointwise, pairwise, and listwise objectives to refine the model's ability to discern subtle semantic distinctions. Crucially, we propose an innovative Parallel Slice Ranking Reward (PSRR) mechanism that computes ranking rewards in parallel slices, where each slice comprises same-indexed completions from different samples. This provides a precise, differentiated learning signal for each individual completion, enabling granular credit assignment and effective optimization. On the official C-STS benchmark, PoLi-RL achieves a Spearman correlation coefficient of 48.18, establishing a new SOTA for the cross-encoder architecture. As the first work to successfully apply RL to C-STS, our study introduces a powerful and precise paradigm for training LLMs on complex, ranking-based conditional judgment tasks.

This past Friday, I asked John Mueller of Google a bit more on if and how Google may use machine learning for adjusting the weights of various ranking signals. The short answer is, Google may or may not do this, depending on the specific ranking signal. But keep in mind, I narrowed the question specifically to if and how Google may use machine learning for adjusting the weights of individual ranking signals. It may be also the case machine learning is used amongst multiple ranking signals or to maybe even create new ones based on the query. It is hard to know for sure.

Regular readers will likely wonder what more I could have to say about machine learning (ML) in search, after having written How Machine Learning In Search Works just a few months ago. Let me assure you, this article is different. Today you won't be reading the ramblings of an SEO professional who fancies himself reasonably informed in how machine learning works as it's related to search. Instead, we'll be turning the tables and learning about search implementations from the perspective of a machine learning expert. This article outlines and hopefully expands on some of the core concepts discussed in an amazing interview with fellow Search Engine Journal contributor Jason Barnard and Dan Fagella of Emerj.

We are seeing more references to machine learning in how Google is ranking pages and other documents in search results. That seems to be a direction that will leave what we know as traditional, or old school signals that are referred to as ranking signals behind. It's still worth considering some of those older ranking signals because they may play a role in how things are ranked. As I was going through a new patent application from Google on ranking image search results, I decided that it was worth including what I used to look at when trying to rank images. Images can rank highly in image search, and they can also help pages that they appear upon rank higher in organic web results, because they can help make a page more relevant for the query terms that page may be optimized for.

Have you thought about optimizing the meta descriptions for your website? Do you know how they impact SEO? Meta descriptions provide short summaries of websites and pages to readers. They give people an idea of what to expect from a website, and if you use them properly, you can get more clicks and conversions too. Let's take a closer look at how to write great meta descriptions for SEO and get more clicks for your website. The meta description is a short piece of text that describes your web page.

Getting Over Ranking Factor Studies in the Era of Machine Learning September 25, 2018 Posted by Mordy Oberstein Just admit it, SEO is scary. Between the inherent complexity of what we do and Google not exactly being the epitome of clarity, the ground that is doing SEO can be a bit shaky at times. That's pretty much why we're obsessed with what works and what doesn't work and are vigilantly on the lookout for content that offers a bit of light at the end of the tunnel. In the not too distant past, I wrote a piece highlighting how machine learning has impacted rank volatility (in that rank is considerably more volatile). At the time, we touched on what machine learning means for understanding how ranking works and how the process directly influences rank. Here, we'll get into the nitty-gritty of it all by analyzing the holy of holies of optimization information, ranking factor studies, particularly niche ranking studies by asking one very simple question .... Do ranking factors studies still apply in a world where machine learning and intent reign supreme, and if so, to what extent? Recap of Machine Learning's Impact on Rank The increase in rank volatility aside, in what for all intents and purposes was "Part I" of this post we discussed how machine learning impacts rank qualitatively, i.e., what rank "looks like" as a result of RankBrain and the like. Since I'm a nice guy, let me recap (and expand on) what we said there so that you don't have to comb through the last piece trying to glue together all of the pieces to the puzzle. Machine Learning Sets Site Proportions In serving up results that align to user intent, Google uses machine learning to determine the proportion of sites to meet that intent or those intents. OK, Mordy, say that in English, please?! If you'll remember, in the last post I took a very straightforward search term, buy car insurance, and showed that Google sees two (or really more than two) intents embedded in that phrase: to buy an actual insurance policy and to get information about doing just that. How should Google handle these two intents?

Recently, Google made another one of its fascinating yet painfully ambiguous announcements: the company is now relying on the power of an artificial intelligence system known as RankBrain to monitor its search results, make progress where necessary, and ultimately guide the development of their core search algorithm. In fact, over the past several months, a "substantial percentage" of all Google queries were handled by RankBrain itself. The average user (i.e., people like you and me) hasn't noticed anything especially different, but according to Google's plan and current available data, this move to RankBrain will make the web an easier, better place to search. That being said, if we haven't noticed much of a difference, why is RankBrain such a big deal? What is it actually doing, and why is that relevant to the average user?

Will artificial intelligence (AI) put marketers out of work? It's like if everyone 150 years ago was asking: "Will the tractor put farmers out of work?" Of course, John Deere didn't put farmers out of business; better tools just made them more efficient and better able to scale. Granted, the tractor did reduce the demand for horses and farmhands. So, no, AI will not put you out of work…as long as your work is creative, innovative and intelligent.

Google uses a machine-learning artificial intelligence system called "RankBrain" to help sort through its search results. Wondering how that works and fits in with Google's overall ranking system? Here's what we know about RankBrain. The information covered below comes from three original sources and has been updated over time, with notes where updates have happened. First is the Bloomberg story that broke the news about RankBrain (See also our write-up of it).

In the past couple of weeks there has been a reinvigorated fervor surrounding artificial intelligence, with "AIO" (Artificial Intelligence Optimization) rearing its head on agency websites and blogs. HTTPS and mobile first seem to be cooling as topics, so attention is turning to RankBrain. The reality of this however is that artificial intelligence optimization is seemingly a paradoxical notion. If we imagine that Google is a child, when the child goes to school and reads a book, we want the child to learn and understand the information in that book. If the book isn't "optimized" for the child to learn – structured information, images, engaging, positive user experience etc. – then the child won't learn or understand the content.