Abstract:

[Purpose/Significance] The existing AI library research is mostly limited to isolated technologies such as recommendation algorithms and chatbots, and lacks a systematic explanation of the overall mechanism of the service model. From the perspective of resource bricolage theory, based on the realistic constraints of capital, technology, and talents, this study constructed an integrated theoretical framework and revealed the elements, hierarchical relationships, and dynamic evolution logic of the library's personalized intelligent service. On this basis, the core issue of "how to realize artificial intelligence-driven innovation in libraries with limited resources" is analyzed to make up for the shortcomings of existing research in applying bricolage theory to the field of personalized intelligent services. [Method/Process] This study adopted a combination of multiple-case study and grounded theory. According to the principle of maximum difference, seven cases were selected: the National Library of Singapore, the National Library of Australia, Stanford University Library, Helsinki Central Library in Finland, an Ethiopian university library, the Library of Congress of Chile, and the Digital Library of Jao College of India, covering three types (public, university, and professional libraries) and spanning six continents (Asia, Oceania, North America, Europe, Africa, and South America). Materials were sourced from policy documents, academic papers, research reports, and news reports. Non-Chinese materials were subjected to two-way back translation to ensure semantic consistency. NVivo 15 software was used for three-level coding: in the open coding stage, initial concepts were extracted sentence by sentence, yielding 128 initial concepts, which were merged into 37 initial categories; in the axial coding stage, the 37 categories were summarized into 12 subcategories, which were further refined into four main categories: value goals, technology empowerment, scenario application, and support and guarantee; in the selective coding stage, the two-way interactive logic of "target traction-foundational support" among the four layers was established, and seven types of resource bricolage were identified: experience-driven, institution-driven, capability-driven, and organization-driven, demand-oriented, technology-deep integration, and agile tool-based. Two articles were reserved for theoretical saturation testing, confirming that the core categories reached saturation. [Results/Conclusions] We have constructed a "four-layer dynamic model driven by resource bricolage," and support the bottom-up support layer (data governance, librarian ability, organizational culture, and ethical norms) as a bricolage resource library; the technical enabling layer (data intelligence and algorithm intelligence) realizes the creative reconstruction of resources; the scene application layer (resource discovery, virtual consulting, scientific research service, and adaptive experience) is the value transformation interface; the value target layer (precise, efficient, immersive, and inclusive) provides direction traction. The four layers constitute a recursive cycle consisting of target traction, resource activation, organizational improvisation, resource reconstruction, scenario realization, and capability precipitation, thereby promoting the spiral upward development of service capability. We define the three-stage path: basic construction (resources will be available, suitable for grassroots libraries), in-depth integration (resource reconstruction, suitable for pilot libraries), and ecological innovation (system bricolage, suitable for leading libraries). This paper contributes by introducing bricolage theory, constructing the structure-process integration framework, and proposing a three-stage implementation roadmap. Its limitations are that it relies on second-hand information and lacks first-hand interviews, as well as universality for an empirical test. In the future, a quantitative evaluation of librarian AI ability modeling and deep impact research on generative AI can be conducted.

Key words: artificial intelligence, resource bricolage, smart personalized service, smart library, personalized service, service model, multi-case study