Recommending blue ocean technologies for subcontractors: A framework based on business reports of prime contractors and patents

Subcontractors depend heavily on their prime contractor and thus find it very risky to enter a new business on their own. This study proposes a framework for these subcontractors to develop blue ocean technologies related to their prime contractor. First, the primary technologies predicted to be promising are extracted from the business reports of the prime contractor. Sub-technologies are then selected through a patent-based search using keywords and International Patent Classification codes of the primary technologies. From them, blue ocean technologies are proposed by optimizing the weighted mean of the min-max normalized market value, degree of competition in the technology market, and subcontractors’ potential technological capabilities for each sub-technology. This study shows that subcontractors can enhance their technology competitiveness by finding a low-risk blue ocean technology. Our empirical research on the subcontractors of a semiconductor firm identified technological patent fields for them to pursue. From our framework, subcontractors can identify blue ocean technologies by considering their prime contractor’s future industrial areas and technologies of interest as well as their own technological capabilities. Furthermore, the prime contractors can gain the synergy effect of technology expansion through cooperation.


Introduction
Subcontracting, also called outsourcing, is an efficient and economical method for prime contractors to access the resources they need [1,2]. Subcontractors are useful partners for prime contractors to diversify their market risks by reducing their operating costs and increasing their competitive advantage [3]. Subcontractors perform tasks that their prime contractors cannot carry out efficiently by themselves, using their unique technology in a more cost-effective manner [4]. Thus, subcontracting is an important business option for prime contractors to flexibly respond to dynamically changing markets and save money on hiring, training, and codes of the promising technologies. In order to attract the prime contractor, the subcontractor needs to be equipped with valuable technologies which do not have many competitors. We call them blue ocean technologies. To select blue ocean technologies among the sub-technologies, we consider three patent-based factors: the sub-technologies' market value, sub-technologies' degree of competition in the technology market, and subcontractors' potential technological capability for each sub-technology which represents how similar the subcontractors' technologies are to their prime contractors' technologies in terms of their patents. Blue ocean technologies are selected using an integer programming to maximize the weighted sum of the three factors. This paper is organized as follows. Section 2 presents the related literature. Section 3 describes the framework of the study. Section 4 shows an empirical study applying the proposed framework, and Section 5 discusses our results in relation to the existing studies. Section 6 concludes the paper with a summary.

Literature review
Companies continue to pursue new trends and develop appropriate skills to survive in the market [26]. In this technology-driven economy, with corporate innovation growing rapidly, companies need to decide on technology adoption continuously [27]. The advanced technologies adopted bring significant benefits to companies along with innovation [28,29] and induce the interest of investors and governments [30]. However, the fluctuating technology market makes it impossible for a company to perform all its tasks using its internal R&D alone [22]. Subcontractors can solve this problem by working as partners to their prime contractors in the technology market. Subcontracting is a commercial contract wherein a subcontractor carries out the prime contractor's responsibilities and duties [31]. This has been in use for a long time, with wider implementation in recent days [32]. The advantage of subcontracting is that subcontractors are specialized in a few tasks, and can thus deal with the tasks of their prime

PLOS ONE
contractors more efficiently and rapidly [1]. As the value of subcontracting increases, we need to pay attention to technological development or innovation to subcontractors.
Several studies have examined various aspects of subcontracting because this is an important issue. They mainly focused on the subcontractor's bidding [33,34], the management relations between subcontractors and prime contractors [15], and investment from the perspective of the prime contractor [35][36][37]. Since innovation is important for a continuously successful business, in addition to efficiency, some other factors such as the subcontractors' technological knowledge or advanced technology have also become significant issues in subcontracting [38]. In fact, the more innovative and fundamental the subcontractors' technologies are, the less is the subcontractors' uncertainty and the more positive is the development in their relationship with their prime contractors [24]. Furthermore, subcontractors adopting high technology levels evolve by developing their technological competencies [38]. Sturgeon [39] argued that subcontractors play a more leading technological role than prime contractors. Thus, subcontractors need to invest in technological innovation [40] and improve their technological competitiveness by developing additional resources [41]. However, subcontractors prefer familiar tasks instead and hamper their innovation, unaware of their potential capabilities [19,20]. Furthermore, subcontractors usually tend to rely heavily on their prime contractors [16,18,41] and take no initiative on their own technology development. Therefore, subcontractors need to be motivated to develop new technologies, but it is hard to find studies which are concerned with subcontractors' technological development. Cao and Wang [24] argued that technological innovation of subcontractors has a positive correlation with the relationship between subcontractors and their prime contractors. Pardo and Rama [21] argued that a manufacturing factory with experience in subcontracting adopts a higher level of technology than those which have not. However, even if these studies dealt with the topic of technological development or technological innovation, they did not take into account the specificity of subcontractors' technological followership to their prime contractors.
As subcontracting is a kind of auction in which a prime contractor selects a small number of subcontractors among many candidates, it can be helpful for subcontractors to use blue ocean strategy (BOS) for winning the subcontracting process and reducing their risk in developing new technologies. BOS is a method to find low-competitive markets and pursue value innovation, differentiation, and low-cost productivity [42]. It shows companies the possibility of more profitable and rapid growth in an innovative and developmental market compared to that in the existing competitive and crowded market [43]. To determine the feasibility and profitability of commercializing new technologies, we need to evaluate them both subjectively and objectively [44]. The former is based on expert knowledge and experience, whereas the latter is based on quantitative evaluation. Therefore, this study recommends that subcontractors choose technologies based on their prime contractor's business reports as a qualitative method and then analyze those technologies quantitatively using patent data. Patent data comprise a collection of the technical capabilities and expertise in new technology [45]. Since patents are recognized as a quantitative R&D investment performance measure, they are widely used as a technological innovation performance measure [46], and they are useful for technology forecasting [47].

Promising technologies and sub-technologies
Our framework proposing technologies for subcontractors involves three stages. First, we find the technologies with good prospects from the "Management Discussion and Analysis (MD&A)" and "Business" sections of the prime contractors' business reports. For selecting promising technologies, two conditions need to be satisfied: (1) the technical term that did not appear in the previous year's business report should appear in the present year's business report, and (2) it should appear in the present or future positive statements. Then we extract U.S. registered patents through keyword search of these technologies for title, abstract and claims by using WIPSON which is a patent database with a high-quality level in Korea containing full-text information of patents registered in patent offices of 12 countries including big 5 and frequently updating them. The patents found are classified by IPC code, and each group based on IPC code is defined as a sub-technology of the promising technology.

Three factors used for technology selection
Second, from the sub-technologies obtained, we calculate three factors: the market values of the sub-technologies, degree of competition of the sub-technologies in the technology market, and potential technological capabilities of the subcontractors for each sub-technology.
The market values of the sub-technology i, a i , are calculated using the patent quality index (PQI), which shows the comprehensive economic and technological value of patents [48]. PQI is strongly linked to the market value of patents [49]. It provides value to the average min-max normalized patent subindex, including forward citations, family size, number of claims, generality index, backward citations, and inverse grant lag. The definition of each patent subindex for PQI is given in Table 1, based on OECD paper [48].
A patent subindex for PQI is forward citation. It is defined as the number of times a patent has been cited five years after the date of its application. However, if the patent data collected are for the latest period, it is highly likely that the patents considered were applied for less than five years from the present date. Thus, for recent patents, the number of forward citations could be zero. Therefore, we use the zero-inflated negative binomial regression of the patents applied within five years of the present date for predicting the forward citations five years after the date of their application. Zero-inflated negative binomial regression is commonly used for predicting forward citations when their number is zero [50]. This model is found to perform better than a Poisson regression model or negative binomial regression model when zero values are over-distributed and the variance is much greater than the mean [51]. Zero-inflated negative binomial regression is a combination of the count and zero-inflated models, with the former following a negative and the latter following a positive binomial distribution. In this modeling, the number of forward citations is used as dependent variable, and the number of backward citations, number of claims, family size, promising technology to which the patent belongs, and classification code are independent variables as they are commonly used in Table 1. Description of patent subindex for PQI.

Forward citations
Number of citations a given patent receives (for calculating PQI, we use child patents applied five years after the date of their parent patent publication)

Family size
Number of patent offices at which a given invention has been protected

Number of claims
Number of claims of the patent, which represents the scope of the patent protection for technology

PLOS ONE
related studies [52]. We calculate the number of days from the application date to the present point as the offset variable.
The degree of competition of the sub-technology i in the technology market, b i , is calculated as the average of the normalized three components, the number of applied patents belonging to the sub-technology, number of companies that applied those patents, and ratio of the number of the sub-technology's patents to the number of patents with the same classification code.
The potential technological capability of subcontractor j for each sub-technology i, c ij , is calculated using a cosine similarity between the subcontractor's patent document abstracts and those of individual sub-technology, summarized using the term frequency-inverse document frequency (TF-IDF) technique. A high similarity shows the possibility of the subcontractors having high technological capability for a particular sub-technology.
TF-IDF considers both how often a word appears and how important each word is in a document [53]. After assigning a weight for each word using the TF-IDF technique, we calculate the cosine similarity between the patent document abstracts of each sub-technology and each technology held by the subcontractors. We select the highest value of the similarities that each patent of each sub-technology has with the patents of a particular subcontractor, and define it as the representative similarity of each patent m belonging to a sub-technology i for subcontractor j, M ðiÞ m; ;j . Then, we define the p-norm value of the representative similarities of all patents belonging to a particular sub-technology i as the potential technological capability c ij that subcontractor j has for the sub-technology i, as follows: where p i is the number of patents belonging to the sub-technology i and M ij ¼ ðM ðiÞ 1; ;j ; M ðiÞ 2 ;j ; . . . ; M ðiÞ p i ;j Þ is a vector consisting of M ðiÞ m ;j for all m of a particular sub-technology i and subcontractor j. The p-norm reflects both the average size and maximum value of the components belonging to M ij . A comparison of the potential technological capability that subcontractor j can have for each sub-technology i shows that the higher the c ij , the higher is subcontractor j judged to have the technological capability potential for the particular subtechnology i.

Optimal selection of blue ocean technologies
Finally, using linear programming, we find the optimal solution that maximizes the weighted normalized sum of the three factors for the recommended technology selection. The weight of each factor is determined by the business strategy or circumstances of each subcontractor. The objective function and constraints of individual subcontractor j are as follows: where w a , w b , and w c are the respective weights of each factor, whose sum is 100, and n is the total number of sub-technologies subcontractor j plans to choose. As we attempt to find blue ocean technologies, we subtract the competition score b i from 1 and define it as an inverse competition score. We consider the exponential on 1−b i because its difference is small.

Empirical study
In this section, we apply the aforementioned methodology to an actual case of a semiconductor industry. This industry is known as an intellectual property intensive, heavily relying on R&D with rapid technological changes and overall increase in patenting [54,55]. We choose SK Hynix Inc., the world's third-largest company in terms of revenue among all semiconductor vendors in 2018 [56], as the prime contractor, and some of its subcontractors.

Promising technologies and their sub-technologies
From the business reports for 2018 and 2019 issued by SK Hynix Inc., we found the company's promising technologies by identifying the technological terms that were not present in the 2018 business report but appeared in the 2019 business report. They are high bandwidth memory, high-end graphics card, foldable display, and high-resolution display. We then extracted the U.S. patents registered from 2017 to 2019 using the keywords of the promising technologies as of June 19, 2020. Thus, we got 245 patents; these have 110 main IPC codes. We classified the patents by their main IPC codes, each representing a sub-technology. The results are presented in Table 2.

Market values of sub-technologies
We calculate the PQI values of each sub-technology by the definition given in Table 1. Because the forward citations of patents showed a right skewed distribution and numerous zero values, we employed zero-inflated negative binomial regression to predict the number of patent forward citations for five years from the patent application date.
A zero-inflated negative-binomial model consists of two sub-models: the count model and zero inflated model. For each model, we used the following independent variables: number of backward citations, number of claims, promising technology, and family size. For the nominal variable of promising technology, high resolution display is set as the baseline category. We used the Akaike Information Criterion (AIC) value for variable selection of each model. Table 3 represents the fitted count model with log link, and Table 4 represents the fitted zeroinflation model with logit link for zero forward citation. Table 5 presents the descriptive statistics of the sub-technology PQI values classified by promising technologies. We find that the display sectors generally have larger PQI values than the rest. This indicates that more sub-technologies generally have higher market values in display sectors than in other sectors on average.

Degree of competition between sub-technologies in the technology market
We calculate the competition score of each sub-technology. To summarize the results, Table 6 shows the descriptive statistics of the sub-technologies' competition scores classified by promising technologies.

Potential technological capabilities of subcontractors for each subtechnology
We chose three subcontractors of SK Hynix Inc.-Daeduck Electronics, Soulbrain, and UniTest. They are selected as they are financially stable and superior subcontractors that are suitable for challenging new technology adoption or expansion. Daeduck Electronics is an electric and electronics company mainly dealing with the manufacture of laminated printed circuit board plates. Soulbrain is a semiconductor and display company mainly engaged in chemical manufacturing. UniTest is a machinery and equipment company mainly manufacturing machines for semiconductors. These companies focus on domestic business, and their patents were disclosed or granted by the Korean Intellectual Property Office (KIPO) from 2017 to 2019. The p-norm of maximum similarity between the patents of each sub-technology and subcontractor indicates the potential technological capability of each subcontractor for each sub-technology. Table 7 lists the descriptive statistics of the potential technological capabilities classified by promising technology. Fig 3 displays the potential technological capability of each subcontractor for each sub-technology. Daeduck Electronics shows the highest technological capability in the high bandwidth memory sector, whereas both Soulbrain and UniTest show the highest technological capability in the foldable display sector. Daeduck Electronics has the highest technological capability in H01L-023/02 of the high bandwidth memory sector, Soulbrain in H05K-005/00 of the foldable display sector, and UniTest in H05K-001/02 of the foldable display sector, with 0.326, 0.537, and 0.435, respectively.
To evaluate the sub-technologies themselves as blue ocean technologies, Fig 4 shows a subtechnology map of normalized values for PQI and exponential value of one minus competition score as coordinates.
Most of the sub-technologies, especially those in the high-end graphics card sector, are clustered near a coordinate of (0.25, 1.00). However, the sub-technologies in the foldable display sector are also widespread in areas where the normalized PQI is relatively high.

Optimal selection of blue ocean technologies
We carry out a sensitivity analysis on how the recommended sub-technologies change by controlling the objective functional weight in linear programming. We find the optimal solution

PLOS ONE
by setting each weight to at least 20% of the total weight and changing the weight in 10% units of the total weight. Table 8 shows the recommended sub-technologies as optimal solutions following the change in weight, which resulted in change in the recommended sub-technologies for subcontractors. Therefore, we find that the decisions on technology development and expansion can vary depending on the strategies the subcontractors adopt. If the subcontractor pursues mainstream technologies rather than self-potential capability, it would provide a higher weight to the market value and inverse competition scores. In contrast, if the subcontractor focuses more on its main technology, it would grant its potential technological capability a higher weight. The technology recommended for all the three subcontractors is G06F-003/042 (digitizers such as those for touch screens or touch pads, characterized by opto-electronic means), which has the highest PQI and is therefore frequently recommended as top-tier technology, especially when we give high weights to the market value or competition score. This means that we can consider the technology itself as a blue ocean, regardless of the potential technological capability of the subcontractors. In fact, digitizer-related technology shows higher marketability in its similar technology group, as the average number of patent family countries with IPC code starting with G06F registered in the first half of 2020 is 2.63, compared to 3.98 for patents with IPC code G06F-003/042.
For Daeduck Electronics, when the lowest weight, 20%, is given to the market value of technologies, H01L-023/02 (semiconductor containers or seals or other solid-state devices over which H01L 23/12, H01L 23/34, H01L 23/48, and H01L 23/552 take precedence) is mostly ranked the highest. This technology tends to show an improvement in ranking as the weight of the market value decreases. When the highest weight, 60%, is given to technological capability, H05K-001/11 (printed elements for providing electric connections to or between printed circuits) is the newly detected technology recommended. Regardless of weight, H05K-001/02 (details of printed circuits) is mostly in the list of the top three recommended technologies.
For Soulbrain, when the weight of the market value is decreased, H05K-005/00 (casings, cabinets, or drawers for electric apparatus) and H01J-001/60 (incandescent screens on or from which an image or pattern is formed, picked up, converted, or stored) are prominently detected in the top three rankings, indicating an improvement in rankings. H05K-005/00 and

PLOS ONE
H05K-005/02 (details of casings, cabinets, or drawers for electric apparatus) are always in the top three ranks regardless of weight. For UniTest, when the weight of the market value of technologies is not the highest, G06F-001/26 (power supply means, such as regulation thereof for memories G11C) is prominently detected in the top three rankings. In particular, when the lowest weight, 20%, is given to the market value of technologies, G06F-001/32 (power supply means for saving power) is the newly detected recommended technology in the top three ranks. H05K-001/02 (details of printed circuits) is always in the top three ranks regardless of weight.

Discussion
Through the empirical study, we identified recommended technologies for individual subcontractors with changes in the weights of three factors. It backs up the studies that decisions on the expansion or development of technologies can vary depending on the subcontractors' strategy and direction [57,58]. There are two cases: (1) a commonly recommended technology for all the subcontractors and (2) technologies recommended differently for each subcontractor. The former supports previous studies that it is important to track and identify the

PLOS ONE
technology trend for business survival [59,60]. It can explain why different subcontractors can be recommended with the same technology in common. The latter supports existing studies that a new technology needs to be chosen not only by the value of the technology itself, but also by the subcontractor's current technology level and potential technological capability [1,21,26,41]. This study provides insights to the existing studies [15,16,18], and the existing problem in subcontracting where win-win relationship between subcontractors and their prime  contractors has not been considered carefully. Previous studies have focused on subcontractors being chosen by prime contractors while have not fully considered subcontractors' technological followership to their prime contractors. The contribution of our paper is to consider their win-win relationship with reflecting the subcontractors' technological followership to their prime contractors, which makes the risk of subcontractors' technology development lower. This is why our study contributed to the existing problem in subcontracting.

Conclusions
This study proposed a framework for subcontractors so as to choose blue ocean technologies based on a win-win relationship with their prime contractors. In this framework, one finds promising technologies from the business reports of the prime contractor and classifies the IPC codes of patents searched by keywords of the promising technologies to identify sub-technologies. Then three factors are used to optimize the sub-technology-selection process: market values of the sub-technologies, degree of competition of the sub-technologies in the technology market, and potential technological capabilities of the subcontractors for each sub-technology. These factors are measured by the PQI; average of the normalized number of applied patents of the sub-technology, number of companies that applied those patents, and ratio of number of sub-technology patents to number of patents with the same classification code; and the cosine similarity with the TF-IDF technique, respectively. The proposed framework is applied to the case of the subcontractors of SK Hynix Inc.: Daeduck Electronics, Soulbrain, and UniTest.
As a result, we found out following. Digitizer-related technology such as G06F-003/042 was commonly recommended to those subcontractors, as it has a high score as a blue ocean technology. On the other hand, different technologies were recommended depending on the potential technological capability of each subcontractor by increasing the weight of the technological capability factor. Technologies related to containers or seals of semiconductor (H01L-023/02), incandescent screens (H01J-001/60), power supply means (G06F-001/26) were respectively suggested to Daeduck Electronics, Soulbrain, and UniTest as a differentiation strategy.
This study used a combination of qualitative and quantitative approaches to help the subcontractors decide on developing new technologies and expanding their existing technologies, which benefit both subcontractors and their prime contractors. Due to the imbalance of power between prime and subcontractors, subcontractors have not had initiative to innovate themselves but had to rely on the prime contractors' needs. However, we suggest the three criteria (market value of technology, degree of competition in the market, and the suitability of subcontractors to develop promising technologies) for subcontractors to develop blue ocean technologies that can benefit both the prime and subcontractors. Subcontractors may adjust the weight of three factors based on common interest of them and their prime contractors. Externally, our framework fully reflects the subcontractors' business relationship with their prime contractors and reduces their risks of technology development. Internally, the framework recommends technologies considering the subcontractors' technological capabilities. Therefore, by applying the methodology of this study, subcontractors can have a positive impact on their own as well as their prime contractors' innovation by developing and expanding the recommended technologies. This study thus enables subcontractors to have a long-term technological competitive edge and coexist with their prime contractor as well as positively contribute as technology providers to the supply chain.
However, the study has following limitations. In the empirical study, there was an issue of small sample size, as some promising technologies do not have many patents. In predicting forward citations, one of subindices for PQI, we could not include more diverse independent variables. Extension to these areas is left for further research.
Supporting information S1 Dataset. Patent dataset for this paper.