Putting a number on novelty
This post dives into the measurement of invention novelty, explaining how researchers have attempted “to quantify the unprecedented.”
At first glance, the idea of measuring the degree of novelty may seem odd to patent attorneys and examiners. Novelty is a patentability requirement: an invention is either novel, or it is not. Any patent application that survives examination must, therefore, clear that threshold—and some researchers have sought to model that threshold [1]. But the economic notion of novelty is different. Just as inventions can be more or less “important,” they can also depart from existing technology to varying degrees. In that sense, novelty is not a binary outcome but a continuous one: some inventions are simply more novel than others.
Measuring invention novelty is central to the economics of innovation. Macroeconomists care about novelty because it helps distinguish incremental improvements from technological breakthroughs. Incremental inventions refine existing technologies and typically generate modest economic gains. By contrast, highly novel inventions can open new technological trajectories and eventually create new products, markets, and industries. Microeconomists, in turn, seek to measure novelty independently of eventual success, thereby avoiding survivor bias. Observing genuinely novel inventions that fail offers a window into the sources of failure, such as weak market power or financing constraints. Finally, management scholars use novelty measures to study the conditions that foster breakthrough inventions. For example, they shed light on whether team structures, funding mechanisms, organizations, or policies are more conducive to breakthrough advances than to routine innovation.
Novelty is a latent construct
A strand of the literature distinguishes novelty according to where an innovation is new. The Oslo Manual, which sets international guidelines for defining and measuring innovation, distinguishes innovations that are new to the firm, new to the market, and new to the world. For patented inventions, novelty is de facto new to the world because patent systems typically apply a worldwide novelty requirement. Hence, the question is not whether an invention is new to the world, but how new to the world it is.
Unfortunately, there is no objective, universally accepted measure of invention novelty. In the economics literature, technological novelty is what we call a latent construct: like happiness or intelligence, it is a concept we can recognize but cannot observe or measure directly. Scholars have, therefore, developed a variety of patent-based indicators intended to capture it. To validate these measures, they often test whether the indicators identify patents associated with independently recognized breakthrough inventions. Breakthroughs can be found in Nobel Prize-winning inventions, inventions by National Inventor Hall of Fame inductees, or winners of the R&D 100 Awards. A complementary strategy is to construct a benchmark sample at the opposite end of the spectrum: patent applications that examiners rejected under statutory provisions for lack of novelty [2].
Scholars have approached the measurement of novelty from several angles. Some examine an invention’s technological components, asking whether they are combined in unprecedented ways or whether they are themselves new. Others examine an invention’s knowledge lineage, asking how different its intellectual ancestry is from that of earlier patents. More recent approaches compare the patent texts, measuring how different an invention’s technical content is from what came before. Finally, a separate family of indicators looks explicitly at subsequent inventions, asking how strongly a patent shapes what comes afterward.
The recombination perspective
As far as I know, Lee Fleming was the first to develop quantitative indicators that combine patent classification codes to measure the novelty of technological recombinations. He used USPC codes as technological components and calculated how frequently they had been combined in the past. In this “recombination” perspective, a novel invention recombines pre-existing technological components in an unprecedented pairwise configuration that has rarely, if ever, been observed in the historical patent record. Using a food analogy, this approach is similar to pairing common ingredients, like strawberries and balsamic vinegar; the metric scores high because the combination is statistically rare, even if the individual ingredients are common.
Building on the recombination perspective, researchers have made key statistical improvements to Fleming’s original metrics. While Brian Uzzi, Ben Jones, and colleagues first introduced this approach to measure the atypicality of journal citation pairs in scientific papers, Daniel Kim and colleagues adapted the method directly to patent technology classification codes. These scholars recognize that raw frequency does not equal typicality: two technology codes might never be combined simply because both are extremely rare. To address this issue, they ask whether a pair of technology codes appears together more or less often than we would expect by chance, given how common each code is overall (using a null model). This method allows them to distinguish combinations that are rare by accident (like strawberries and balsamic vinegar) from those that are genuinely rare (like caviar and truffles).
Another way to measure novelty in the recombination tradition is to check whether an invention introduces knowledge from an external domain that has not previously been brought into that specific technological field. Dennis Verhoeven, Reinhilde Veugelers and colleagues look for novel pairwise crossings, flagging when a patent assigned to a certain technology class cites a prior patent class or a broad scientific field that has not been cited by patents in that focal class before. Pursuing the culinary analogy, this measure is equivalent to a pastry chef borrowing a standard spice from savory cuisine and crossing that boundary to add it to a sweet dessert recipe for the very first time.
The knowledge lineage perspective
Kristina Dahlin and Dean Behrens operationalize novelty using citation similarity. Their measure calculates the overlap (or similarity) of backward citations between a focal patent and all patents from previous years; a lower overlap score indicates higher relative novelty. In this “knowledge lineage” perspective, a novel invention has a backward citation structure that is highly dissimilar to and has minimal reference overlap with the patents that preceded it in the same technological field. Imagine that every recipe lists the earlier recipes that inspired it. Looking backward at the technical lineages a dessert draws from, a new dessert does not cite traditional French baking texts. Instead, its technical foundations are borrowed from ancient Mesoamerican chocolate fermentation and modernist molecular gastronomy centrifugation. Because this dessert’s “recipe lineage” has minimal reference overlap with the standard pastries that came before it, the dish achieves a high novelty score.
Dahlin and Behrens’ original metric required calculating dyadic citation-overlap scores against the entire population of patents, which made large-scale implementations across multiple technological fields computationally expensive. Sam Arts, Bart Van Looy and colleagues resolved this computational bottleneck by defining a more targeted comparison group. They compared each focal patent’s citation profile only to other patents that shared at least one 3-digit technology class with the focal patent.
The textual perspective
Scholars are gradually moving away from technology classification and citation metadata. The former is a coarse measure of knowledge origins, and the latter is prone to strategic manipulation, examiner additions, or omission errors. Several scholars have translated the “dissimilarity to prior art” concept into the unstructured text space to obtain measures of semantic similarity [3,4,5]. For example, Sam Arts and colleagues extract unique keywords (as well as bigrams and trigrams) from the patent text. They measure novelty by calculating the number of these new terms and pairwise keyword combinations introduced by a patent for the first time in history, as well as the focal patent’s overall semantic distance to all patents filed in the preceding five years. In our kitchen analogy, instead of looking only at broad ingredient categories or which earlier recipes are cited, this approach reads the recipe. A novel dish is one that introduces novel terms (ingredients, cooking instructions, etc.) or novel combinations of these terms.
More recent studies use text embeddings to measure technological similarity. Text embeddings convert patent text into numerical vectors that capture meaning and context, rather than simply counting shared words. Patents describing similar technologies cluster together in this numerical space, even when they use different terminology, while those describing very different technologies are farther apart. The food analogy is a giant culinary map on which similar recipes sit close together. “Slow-cook the beef” and “braise the meat for several hours,” for example, would be recognized as describing similar techniques despite using different words. Leveraging this approach, Daniel Hain and colleagues identify patents that sit far from other patents. They measured novelty as a lack of similarity to preceding patents. In a parallel advancement, Daeseong Jeon and colleagues found that conventional patents cluster in dense regions of the vector space, while highly novel patents occupy isolated, low-density regions. In culinary terms, the most novel dish is one sitting alone in a largely unexplored corner of the recipe map.
Beyond novelty: disruption
The novelty literature has branched into a stream that focuses on the disruptive nature of patents. Patent novelty is typically measured ex-ante using information available at the time of filing (be it from technology classes, citation metadata, patent text, or a combination of these measures). By contrast, forward-looking measures, which compare a patent with what will come after it, are measures of “impact.”
Russell Funk and Jason Owen-Smith developed a consolidation-disruption index that measures whether subsequent patents cite the new invention alongside its predecessors (consolidating) or instead of them (disruptive). Using a food analogy, a consolidating patent adds a great new side dish that complements the existing menu, while a disruptive patent introduces a main course so revolutionary that future chefs completely stop cooking the older recipes it was based on.
This metric gained prominence when Russell Funk, together with Michael Park and Erin Leahey, applied it, reporting a striking decline in the disruptiveness of patents (and scientific papers) over time. However, subsequent studies have challenged both the measure and its interpretation, citing issues such as data truncation, the secular growth in patent citations, and properties of the index itself [6,7,8,9]. Once researchers account for these issues or modify the index accordingly, the apparent decline in technological disruptiveness becomes substantially weaker. However, some of these criticisms have been rebutted by the original authors’ team.
The disruption controversy illustrates how difficult it is to reduce technological change to a single number. Novelty measurement has come a long way—from patent classes and citation networks to semantic embeddings—but it remains a very active area of research. Patent law may treat novelty as a yes-or-no question; for economists and management scholars, measuring how novel an invention is remains an open question.
If you enjoy evidence-based takes on patents and innovation, join hundreds of readers who receive The Patentist directly in their inbox.
Please cite this post as follows:
de Rassenfosse, G. (2026). Measuring novelty. The Patentist Living Literature Review 15: 1–5. DOI: TBC.