Prior research has shown that the size of the knowledge repository and its interdependencies (i.e., complexity) significantly affect the likelihood of successful knowledge recombination (Kauffman, 1993; Fusillo, 2023; Fleming and Sorenson, 2001). Excessive redundancy in the components of technological knowledge may restrict the range of potential knowledge exploration, inevitably leading to cognitive lock-in (Nooteboom, 2000). In contrast, the diversity of a knowledge base is regarded as a fundamental requirement for long-term technological development (Saviotti, 1996). A series of studies have shown that the expansion of technological knowledge components across different domains within a knowledge base not only broadens the scope of searchable knowledge but also generates more potential technological combinations, thereby increasing the likelihood of successful recombination (Rosenkopf and Nerkar, 2001; Novelli, 2015; Baumann and Siggelkow, 2013). However, greater diversity in the knowledge base also increases the complexity of technological interdependencies. The invention process not only requires managing the substantial noise information resulting from the expanded scope of knowledge search while avoiding the trap of local optima (Dunne and Dougherty, 2016). Choi et al. (2018) observe that scientific knowledge facilitates the comprehension of interdependencies among technologies, the identification of effective technological combinations, and the enhancement of efficiency in complex knowledge recombination. It can be inferred that the more complex the process of knowledge recombination, the greater the likelihood that scientific knowledge being utilized in the invention process. Based on this, we propose the following hypotheses:

H1a: The higher the complexity of knowledge recombination, the greater the likelihood of incorporating scientific knowledge components in the invention process.

Recombining technological knowledge components from disparate domains results in different levels of novelty. Keijl et al. (2016) highlights that inventions derived exclusively from the recombination of knowledge components within local technological fields typically demonstrate lower levels of novelty. Conversely, technological combinations based on distant technological origins tend to exhibit higher degrees of novelty. Related empirical evidence supports the idea that knowledge recombinations exhibiting higher novelty possess greater potential value for future technological development. For example, Kaplan and Vakili (2015), based on an analysis of 2,826 granted patents in the nanotube sector, found that novel knowledge recombinations are not only more likely to form the basis for subsequent technological advancements but also yield greater economic value. However, high novelty knowledge recombinations often come at the cost of reduced familiarity and recognizability (Keijl et al., 2016; Aldrich and Fiol, 1994). Due to limited familiarity and recognizability of technological knowledge components, elucidating the vast array of potential technological combinations and their potential impacts on technical issues proves challenging during extensive searches (Afuah and Tucci, 2012). Moreover, distant technological knowledge components may not provide readily usable building blocks. Therefore, to mitigate the variability and uncertainty inherent in knowledge recombination due to the pursuit of novelty, the invention process is likely to depend more heavily on scientific knowledge. On one hand, scientific reasoning helps inventors identify invention opportunities that are far removed from local technological fields. On the other hand, the predictive function of scientific knowledge can better ascertain potential technological alternatives, thereby avoiding experiments doomed to fail (Balconi et al., 2010). Based on this, this study formulates the following hypotheses:

H2a: The higher the novelty of knowledge recombination, the greater the likelihood of incorporating scientific knowledge components in the invention process.

The foundation of inventive activities may rest upon knowledge bases developed at different times, encompassing both recent and longstanding knowledge. The maturity level of the technological knowledge elements used during the recombination of knowledge can influence the degree of dependence of the invention on established scientific knowledge. Utilizing relatively mature (i.e., older) knowledge enables inventors to gain enhanced insights into the technical attributes and potential uses from industry-specific sources. This facilitates effective evaluation of the merits and demerits of knowledge during the recombination process, thereby reducing the likelihood of technological misapplication and errors (Turner et al., 2013). However, incorporating mature knowledge elements, could result in outdated technological solutions, given that numerous potential combinations of technology might have been previously utilized, thereby leaving less room for valuable technological innovations (Sørensen and Stuart, 2000). In contrast, recent knowledge is well-suited to meet changing requirements, reducing the risk of falling into capability traps and increasing the opportunities for knowledge recombination, which supports the ongoing effectiveness of innovation (Petruzzelli et al., 2018; Levinthal and March, 1993; Ahuja and Morris Lampert, 2001). Nevertheless, compared to well-established knowledge that has been validated over time, the benefits of integrating recent knowledge may be offset by technological uncertainties and higher implementation costs (Heeley and Jacobson, 2008; Petruzzelli and Savino, 2014). Nightingale (1998) notes that scientific knowledge is not merely a tool for generating answers but a means to comprehend how technology functions. This understanding can reduce technological uncertainties and lower the costs of trial and error. Consequently, to maximize the benefits of integrating recent technological knowledge, knowledge recombination may increasingly rely on scientific knowledge. This study proposes the following hypotheses:

H3a: The lower the maturity of the recombined technological knowledge, the greater the likelihood of incorporating scientific knowledge components in the invention process.

Inventive experience encompasses the skills, knowledge, and resources that inventors accumulate through prior activities. Existing research suggests that the inertia of past experiences influences the approaches inventors take to seek technical solutions (Liao et al., 2008). In other words, for invention teams with extensive inventive experience, the inertia of past experiences drives them to create new technological combinations based on their existing knowledge base, rather than seeking new knowledge outside their current field (Wang et al., 2017). Consequently, from this perspective, an inventor’s past experience with science-based inventions enhances the likelihood of utilizing scientific knowledge to address new technological challenges in the future. Furthermore, while inventors may draw upon the accumulated knowledge produced by technological progress, a thorough and accurate understanding of science enables them to search and recombine information effectively (Ardito et al., 2021). Previous successful experiences in the application of scientific knowledge by inventors have been shown to ease the challenges associated with integrating new scientific and technological insights, thereby reducing the time spent on knowledge exploration (Novelli, 2015). Based on this, the following hypotheses are proposed in this study:

H4a: The richer the science-based invention experience of the inventor team, the higher the likelihood of incorporating scientific knowledge components during the invention process.

The breadth of an inventor’s technological knowledge can be defined as all technological knowledge elements they possess, reflecting the diversity of the technological knowledge accumulated (Fleming et al., 2007). Simonton (2003) demonstrated that a rich diversity of cognitive elements available for association enhances the likelihood of identifying unique combinations within these elements. This implies that the broader the scope of technological knowledge, the greater the number of options available to an inventor when addressing new problems. In an investigation targeting inventors in the 3M innovation domain, Boh et al. (2014) employed a mixed- method approach combining qualitative (case study) and quantitative research to investigate how the breadth of inventors’ expertise influences their innovative outcomes. Results indicate that extensive professional knowledge indeed plays a crucial role in enhancing inventors’ capacity to produce more inventions. Furthermore, during the process of knowledge recombination, inventors may opt to utilize their existing knowledge base or seek out new external information to formulate solutions for distinct technological problems. Wang et al. (2017) pointed out that inventors typically initiate their process by exploring the existing knowledge space, employing iterative trial-and- error methods to test various combinations of existing knowledge until a satisfactory solution is found. Therefore, we argue that the broader the technological knowledge possessed by inventors or invention teams, the less likely they are to seek external scientific knowledge. This study proposes the following hypothesis:

H5a: The broader the range of technical knowledge possessed by an inventor team, the lower the likelihood of incorporating scientific knowledge components during the invention process.

Recombination complexity (Complexity). Complexity reflects the diversity of the knowledge base and components required in the process of knowledge recombination (Barbieri et al., 2020). This study employs the Originality index proposed by Trajtenberg et al. (1997) as a proxy for recombination complexity. The formula for its calculation is as follows:

Where K represents a specific technological field, and N_p denotes the total number of distinct technological fields represented in all patent references cited by patent p. This study uses the first four digits of the IPC classification (hereafter referred to as ‘IPC4’) to represent technological fields.Nreferences_p indicates the number of patent references cited by patentp, and Nreferences_pk represents the number of those references that belong to the technological field k. The Originality index reflects the degree to which the cited references of a patent are distributed across various technological fields, with values ranging from 0 to 1. A higher originality index indicates that the focal patent draws from a wide variety of technological domains, suggesting that it is the result of combining knowledge from multiple fields.

Recombination novelty (Novelty). Novelty represents the uniqueness in the process of knowledge recombination, typically reflecting the ‘distance’ between new technologies and prior technological knowledge (Barbieri et al., 2020). This study utilizes the improved radicalness index by Squicciarini et al. (2013) to measure novelty. This index not only characterizes the knowledge gap between the technological categories of the focal patent and those of the patents it cites, but also reflects the extent to which the technological invention combines knowledge components in a novel manner. The formula is as follows:

Where n_p denotes the total number of IPC classification codes in all patent documents cited by the focal patent p. CT_j is the number of IPC4 codes in the cited patentj that differ from those of the focal patent p. This index ranges from 0 to 1, where a higher value signifies greater novelty of the patent.

Knowledge maturity (Maturity). Following the approaches of Jiao (2022), we use the average temporal distance between focal patents and backward citations as an indicator of knowledge maturity. The specific formula used is presented below.

Where N is defined as the number of backward citations to the focal patent, Year_focal represents the application year of the focal patent, and Year_i refers to the application year of the ith patent cited by the focal patent.

This research delineates two dependent variables in accordance with the posed research questions: scientific knowledge integration tendency and scientific knowledge integration intensity.

Scientific knowledge integration tendency (Science) assesses the tendency to integrate scientific knowledge into the technological development process, as measured by the citation of scientific papers by patents. In our study, a patent that cites at least one scientific paper, defined as a ‘science-based invention’, is coded as 1. Conversely, a patent that does not cite any scientific papers, defined as a ’non-science-based invention,’ is coded as 0.

Scientific knowledge integration intensity (science_intensity) refers to the degree to which scientific knowledge contributes to technological patent, as measured by the number of scientific papers cited by patents, using methodologies established in prior research (Harhoff et al., 2003; Sung et al., 2015).

Taking into account potential influencing factors, this study incorporates a set of control variables, including the number of inventors (Num_Inventors), the number of applicants (Num_Applicants), applicant experience with science-based inventions (Num_Applicants_Sci), the number of backward patent citations (Num_PC), the speed of technology accumulation (Tech_Cum_Speed), and the type of applicant.

The variable Num_Applicants_Sci is incorporated because existing studies indicate that the knowledge base of patent applicants influences the trajectory and efficacy of their future technological innovations (Katila adn Ahuja, 2002; Maddala GS, 1987). We measured this by calculating the average number of patents granted that had cited scientific literature prior to the application.

Num_PC represents the count of patent documents that an invention’s patent cites. By controlling for the number of backward citations associated with each invention, we aim to mitigate the extent of technological knowledge dependency, as discussed by Gruber et al. (2013).

Tech_Cum_Speed represents the speed at which a certain technology develops or evolves along its technological trajectory (Lee and Lim, 2001; Guo et al., 2013). The extent of cumulative development across technologies may affect the propensity of that the knowledge reorganization process relies on scientific knowledge (Grupp and Schmoch, 1992; Meyer, 2000). The formula is as follows:

Where n_p represents the cumulative number of the IPC4 classification codes for the focal patent p in its filing year. n₅ indicates the cumulative number of IPC classification codes from five years prior. The greater Tecℎ_Cum_Speed_p is, the faster the rate of accumulation in the technical field to which the invention belongs.

Regarding the type of applicant, referring to the applicant categorization algorithm proposed by Du Plessis et al. (2010), we classify patents into five categories: individual, company, government, university, and hospital, and set five dummy variables.

In addition, we include a set of time, country, and technology field dummies to control for potential time-varying effects, geographical and technological heterogeneity. Specifically, Time is a set of dummy variables for the patent’s granted year. Geo denotes geographical dummies represented by the country of the first inventor listed on a patent. We also include the technological field dummies, IPC, at the 1-digits IPC level (The IPC is a hierarchical system and is divided into sections, classes, subclasses, main groups, and subgroups. The section of IPC includes A, B, C, D, E, F, G, and H, corresponding to eight broad technical fields).

Table 1 presents the descriptive statistics for each variable. The average Science is 0.275, indicating that 27.5% of patents in this study’s sample cited at least one scientific literature. The average scientific_intensity is 3.839, with a maximum value of 3048, representing the highest number of citations to scientific literature by any single patent.