30 Most Common Information Retrieval Aptitude Test Questions

In today's era of data-driven innovation, professional-grade Information Retrieval capabilities are no longer exclusive to library and information professionals but are essential survival tools for every researcher and technical engineer. Facing vast academic literature and technical patents, many fall into the "retrieval equals searching" misconception, relying on intuitive keyword input without realizing that this crude strategy is the root cause of low recall rates and result noise overload. To break this skill barrier and provide a standardized self-assessment system, this article systematically compiles an Information Retrieval Proficiency Test Bank containing 30 high-frequency test points. Integrating the essence of past literature retrieval exams and information literacy competitions, these questions cover database retrieval fundamentals and delve into nesting traps in Boolean logic retrieval exercises, precise truncation control, and interdisciplinary Chinese Library Classification (CLC) code queries. Through the analysis of these typical scenarios, readers will understand why simple logic combinations can lead to the "accidental exclusion" of key literature and how to use controlled vocabulary for precise navigation in heterogeneous databases like EI or Web of Science. This is not merely a knowledge quiz, but a deep review of information acquisition logic designed to help professionals identify knowledge blind spots, upgrade fragmented habits into rigorous systematic intelligence retrieval strategies, and seize the information initiative in research initiation and technical breakthroughs.

Retrieval Language Core: Boolean Logic and Search Syntax

Retrieval Language is the standard interface for interaction between humans and database systems. For engineers, mastering retrieval syntax is not just about memorizing a few operators, but understanding how databases perform set operations on indices through Boolean Logic. When dealing with technical literature, patents, or code repositories, incorrect logical combinations lead to a dual collapse of Recall and Precision.

As the first part of the core competency test, this chapter covers the most fundamental logical operation rules through high-frequency questions. Before entering into complex nesting and truncation techniques, we need to clarify two easily confused concepts and master the underlying truth table of Boolean logic.

Core Concept Distinction: Keyword vs. Subject

Before executing Boolean operations, the nature of the search object must be defined. This is an "invisible error" often made by technical personnel.

• Keyword Retrieval: Based on String Matching. For example, when searching for "Mobile Network", the system looks for the occurrence of these two words in the title, abstract, or full text. The advantage is flexibility; the disadvantage is that it is easily affected by synonyms (such as Cellphone) or spelling differences, leading to missed detections.
• Subject Retrieval: Based on Controlled Vocabulary. For example, in EI Compendex or IEEE Xplore, standardized Thesaurus terms are used for retrieval. Whether the original text writes "Mobile" or "Cellular", as long as they belong to the same subject concept, they can be retrieved.

---

Basic Boolean Logic Truth Value Test (Level 1)

The following questions mainly examine the set operation properties of AND, OR, and NOT. Please note that operator symbols in different databases (such as Web of Science vs. Google) may differ, but the logical essence is consistent.

Q1: In a retrieval strategy, to narrow the retrieval scope and improve precision, which Boolean operator should be prioritized?

Answer: AND (Logical AND)

Technical Analysis:
The AND operator performs a set Intersection operation.

• Logic Model: The search query A AND B only returns records containing both set A and set B.
• Effect: Adds limiting conditions; the result set is necessarily smaller than or equal to any single set.
• Application Scenario: When you need to locate the intersection of specific technology stacks, such as Python AND "Machine Learning".

Q2: If a topic involves the application of "Artificial Intelligence" in the fields of "Medical" or "Education", and there is no restriction on the specific field, how should the basic logic be constructed?

Answer: (Artificial Intelligence AND Medical) OR (Artificial Intelligence AND Education)
Note: A more efficient way of writing usually involves nesting, which will be explained in detail in the next subsection.

Technical Analysis:
The OR operator performs a set Union operation.

• Logic Model: The search query A OR B returns records containing A, containing B, or containing both.
• Effect: Expands the retrieval scope, used to connect synonyms (such as Driverless OR Autonomous) or parallel concepts.
• Visualization: In the logic truth table, as long as one input is True, the result is True.

Q3: When retrieving literature related to "Energy", if you wish to exclude content related to "Nuclear Energy", what operator should be used? What are the risks?

Answer: NOT (Logical NOT); the risk is potentially causing "False Negatives".

Technical Analysis:
The NOT operator performs a set Difference operation.

• Syntax Example: In the EI Database, it is usually written as Solar NOT Nuclear.
• Potential Risk: This is the most dangerous operator. If a high-quality review article is titled "Energy Transition from Nuclear to Solar", although it contains the content you need, it will be directly filtered out by NOT Nuclear because the word "Nuclear" appears in the text.
• Best Practice: Use only when synonym ambiguity is extremely strong (e.g., searching for the Apple brand while using NOT Fruit); it should be used with caution in academic retrieval.

Logical Operation Visualization Quick Reference

After mastering the above basics, the real retrieval challenge often lies in the precedence order of operators and the flexible use of truncation, which directly determine the execution results of complex search queries. The following subsections will conduct specialized testing on these advanced examination points.

Boolean Operator Precedence and Nesting Exercises

When building complex search strings, simply mastering the definitions of Boolean operators is not enough. The root cause of most search failures or cases of excessive noise lies in a misunderstanding of Order of Precedence. Just like arithmetic operations in programming languages, database engines follow a strict execution order when parsing search strings.

The usual default order of precedence is: ()NEAR/SAME (Positional Operators) > NOTANDOR.

The following are three typical high-frequency test scenarios used to examine the mastery of nested logic and execution order.

11. The Mixed Trap of Logic "AND" and "OR"

Question: You need to find literature regarding the "Cultivation" of "Apple" or "Banana". Which of the following search strings can accurately hit the target?

• A. Apple OR Banana AND Cultivation
• B. (Apple OR Banana) AND Cultivation
• C. Apple AND Cultivation OR Banana

Reference Answer: B

Logic Analysis:
This is the most classic logical error. In the default precedence of most databases (such as EI Compendex, IEEE Xplore), AND has higher priority than OR.

• Execution Logic of Option A: The system will first execute Banana AND Cultivation, and then union the result with Apple.
• • Result: You will get all "Banana cultivation" literature, plus all literature regarding "Apple" (regardless of whether it relates to cultivation). This leads to a large number of irrelevant results.
• Execution Logic of Option B: The parentheses force a change in precedence. The system first calculates the union of (Apple OR Banana), then takes the intersection of this set with Cultivation.
• • Result: Accurately hits literature on "Apple cultivation" or "Banana cultivation".

12. Complex Exclusion Logic and Operation Order

Question: In the Web of Science (WOS) database, if the search string is Solar OR Wind AND Energy NOT Nuclear, what is the actual logical order executed by the system?

Reference Answer: First NOT, then AND, finally OR, which is equivalent to Solar OR (Wind AND (Energy NOT Nuclear)).

Logic Analysis:
Default precedence varies across different database platforms, which is an invisible killer in advanced retrieval.

• According to the analysis in Common Information Retrieval Techniques, the default order in Web of Science is NOT > AND > OR.
• Execution Steps:

1. 1. NOT: First exclude records containing Nuclear.
   2. AND: Among the remaining records, find literature containing both Wind and Energy.
   3. OR: Merge the above results with all literature containing Solar.

• Potential Risk: If you originally intended to find "Solar or Wind energy, belonging to the energy field, but excluding nuclear", the above default logic is incorrect (it does not exclude Nuclear from Solar, nor does it restrict Solar to be Energy-related).
• Correction Scheme: Nesting must be used: (Solar OR Wind) AND Energy NOT Nuclear.

13. Multiple Nesting and Concept Combination

Question: An engineer needs to research "Algorithms using Python or Java for Image Processing, but excluding Medical applications". Please write the standard search string.

Reference Answer: ("Image Processing") AND (Python OR Java) NOT Medical

Logic Analysis:
This question examines the comprehensive application of phrase retrieval, Boolean logic, and nesting.

1. Phrase Protection: As a proper noun, it is recommended to use double quotes for Image Processing to force an exact match, preventing it from being split into Image AND Processing.
2. Synonym Expansion: Python and Java are parallel technical means and must be wrapped in () and connected by OR to form a concept group.
3. Logical Order:

• • If written as Image Processing AND Python OR Java without parentheses, the system might parse it as (Image Processing AND Python) OR Java, resulting in the retrieval of all irrelevant literature regarding Java (such as Java backend development).
  • NOT is usually placed at the end or has its scope defined by parentheses to ensure the exclusion condition applies to the entire result set.

Techniques for Using Truncation and Proximity Operators

After mastering the "skeleton" of Boolean logic, advanced search tests usually examine how to handle morphological changes and word relevance through Truncation and Proximity Operators. This part of the questions aims to test the candidate's mastery of fine-tuning tools for Recall and Precision.

The following are the 4 most frequent assessment points in this field:

Q13: In foreign language databases, what is the most efficient way to input if you want to retrieve "computer", "computing", "computation", etc. simultaneously?

• Reference Answer: comput* or comput? (depending on specific database standards, * is generally used)
• Analysis: This is the most typical application scenario of Right Truncation.
• • Principle: Uses a truncation symbol to replace all character variations after the word stem.
  • Trap: Many beginners try to connect all word forms with OR (e.g., computer OR computing OR ...), which is not only inefficient but also prone to missing obscure word forms (e.g., computability).
  • Database Differences: Although * is the most universal infinite truncation symbol (e.g., in Web of Science and EI Engineering Village), some older systems or specific fields may use ? or $. Unless specified otherwise in the exam, * is the default.

Q14: If you need to retrieve "woman" and "women", or "analyzing" and "analysing", which operator should be used?

• Reference Answer: Wildcard, such as wom?n or analy?ing
• Analysis: This tests Internal Truncation or Limited Truncation.
• • Scenario: Mainly used to solve irregular singular/plural changes (man/men) and British/American spelling differences (color/colour, defense/defence).
  • Symbol Standards:
  • • In Web of Science (WOS): ? represents 1 character, $ represents 0 or 1 character (suitable for color/colour), * represents 0 to multiple characters.
    • In EI Compendex: ? represents 1 character, * represents 0 to multiple characters.
  • Technical Point: Using wom?n precisely is better than wom*, because wom* might incorrectly recall irrelevant words like wombat.

Q15: When searching for "Information Retrieval", what is the essential difference between using Information AND Retrieval and Information NEAR Retrieval?

• Reference Answer: NEAR limits the physical distance between two words, resulting in higher relevance; AND only requires both words to appear in the document simultaneously.
• Analysis: This is a core question testing Precision.
• • The Problem with AND: If a 50-page document mentions "Information" on page 1 and "Retrieval" on page 50, they are likely unrelated, but AND will recall it.
  • NEAR (or w/n, adj): Proximity operators force two keywords to be within a specific distance (e.g., NEAR/5 means separated by no more than 5 words). In engineering databases like the EI Database, NEAR (unordered proximity) and ONEAR (ordered proximity) are commonly used to distinguish.

Q16: In the Web of Science (WOS) database, for what special scenario is the operator SAME usually used?

• Reference Answer: Address/Affiliation Search
• Analysis: This is a high-difficulty technical detail question.
• • Definition: The SAME operator requires two keywords to appear in the same sentence or the same field subset.
  • Application: It is crucial when searching for specific departments of specific institutions in WOS. For example, to search for "Yale University Department of Chemistry", one should use Yale Univ SAME Chem. If AND is used, it might retrieve a paper co-authored by "Yale University Department of Physics" and "Harvard University Department of Chemistry" (because both terms appear in the address field but belong to different authors), resulting in false positives.

Technical Summary:
**Truncation (`, `?): Mainly improves **Recall, solving spelling and morphological changes.
* Proximity Operators (NEAR, SAME): Mainly improve Precision, solving contextual relevance.
* Priority: In most systems, the operation priority of proximity operators is higher than Boolean logic operators (NEARAND), but when constructing complex search queries, it is always recommended to use parentheses to clarify the logical hierarchy.

Databases and Evaluation Metrics: SCI, EI, and JCR Quartiles

In information literacy tests at universities and research institutions, understanding the nature of databases and mastering academic evaluation metrics are mandatory content. This section of questions usually tests whether candidates know "where to find what" and "how to evaluate literature quality." The focus is on distinguishing between Citation Databases and Full-text Databases, as well as understanding core concepts such as SCI, EI, JCR quartiles, and Impact Factor (IF).

Below are classic test questions and analyses regarding database characteristics and evaluation metrics:

Core Concept Differentiation: Abstract Databases vs. Full-text Databases

Q6. When conducting a literature survey, what is the primary function of databases like Web of Science (SCI) and EI Compendex?

• A. Directly providing PDF full-text downloads for all literature
• B. Providing bibliographic records, abstracts, and citation indexing of literature, but not necessarily directly storing the full text
• C. Only used for querying the publication year of journals
• D. Specifically used for retrieving Chinese dissertations

Answer: B

Analysis: This is an extremely common misconception. EI Compendex and ISI Web of Knowledge (including SCI) are essentially Abstract & Citation Databases. Their core value lies in indexing metadata (titles, abstracts, authors, citation relationships) of high-quality global literature, rather than directly storing full-text files. Although they usually provide "Full Text Buttons" or DOI redirection functions, actual full-text downloads typically need to be completed by jumping to Full-text Databases such as Elsevier (ScienceDirect), IEEE Xplore, or Springer.

Q7. When you need to find the most comprehensive high-quality non-patent literature in the field of Mechanical Engineering, which database is usually the first choice?

• A. PubMed
• B. EI Compendex (Engineering Village)
• C. CSSCI
• D. OVID

Answer: B

Analysis: Based on subject coverage, Ei Compendex is the preferred solution for scientific research in the engineering field. It specifically indexes journals and conference papers in engineering technology fields. In contrast, PubMed focuses on biomedicine, and CSSCI is the Chinese Social Sciences Citation Index. Understanding the "subject expertise" of each database is the prerequisite for efficient retrieval.

Academic Evaluation Metrics: IF, H-index, and JCR Quartiles

Q8. The "Impact Factor (IF)" of a journal is an important indicator for measuring its academic influence. What is the official publishing source of the IF?

• A. Google Scholar
• B. Scopus
• C. Journal Citation Reports (JCR)
• D. China National Knowledge Infrastructure (CNKI)

Answer: C

Analysis: The Impact Factor (IF) is a metric owned by Clarivate, published in Journal Citation Reports (JCR). Although other platforms like Scopus have their own metrics (such as CiteScore), the "Impact Factor" in the narrow sense specifically refers to the version published by JCR. Candidates should note that the Web of Science database itself is stricter and serves as the data source foundation for the IF.

Q9. In JCR quartiles, when all journals in a specific subject field are ranked in descending order by Impact Factor, what percentage of top-ranking journals does Q1 (Quartile 1) refer to?

• A. Top 5%
• B. Top 25%
• C. Top 50%
• D. Bottom 25%

Answer: B

Analysis: JCR classifies journals by subject and then divides them equally into four parts based on the Impact Factor. Q1 represents the top 25% of top-tier journals, Q2 is 25%-50%, and so on. This differs from the CAS (Chinese Academy of Sciences) partition (pyramid-shaped division, where Q1 is only about the top 5%), so during the exam, one must check whether the question asks about "JCR quartiles" or "CAS partitions."

Q10. What is the H-index, and what is it primarily used to evaluate?

• A. Evaluating the circulation of a journal
• B. Comprehensively evaluating the quantity and quality of academic output of a scholar (or institution)
• C. Calculating the number of times a paper has been read
• D. Counting the indexing speed of a database

Answer: B

Analysis: The H-index indicates that a scholar has $h$ papers that have each been cited at least $h$ times. It is a hybrid metric that examines both publication quantity (output) and citation count (quality/impact), effectively avoiding evaluation biases such as "one highly cited paper masking overall mediocrity" or "a large volume of low-quality publications." In the "Analyze Results" function of Web of Science and Scopus, this metric is often used to identify high-impact experts in a certain field.

How to Use Citation Index for Retrieval

The Citation Index is a core topic in advanced information retrieval tests. It breaks through the limitations of traditional keyword retrieval and can trace the lineage of scientific development through "citation relationships between articles." Below are the most common practical test questions and analyses in this field, focusing on the ability to use citation networks for "snowball" retrieval.

Q1: Given a seminal paper published in 2010, how can you quickly find the latest research results in this field in 2024?

• Answer: Use Cited Reference Search or click the Times Cited link in the database.
• Analysis:
  Traditional keyword retrieval usually only finds literature published before the paper (i.e., its references, Backward Chaining). To find subsequent developments, corrections, or latest applications of the theory, one must use the citation index to trace forward (Forward Chaining). In Web of Science or Scopus, clicking the "Times Cited" link lists all subsequent articles that have cited this 2010 paper. This method helps researchers accurately locate the latest relevant research even when they are unsure if the latest terminology (Keywords) has changed.

Q2: When the number of search results is huge, how can "Times Cited" be used as a quality filter?

• Answer: Sort search results by Times Cited in descending order and prioritize reading highly cited literature.
• Analysis:
  Citation frequency is an important objective indicator for measuring academic impact. High citation counts usually imply that the research has high recognition, validity, or has sparked widespread discussion within the field. According to a relevant tutorial by Tsinghua University Library, using citation links in Web of Science allows one not only to view specific citing documents but also to combine with analysis tools (Analyze Results) to quickly identify core reviews or leading teams in the field.

Q3: If a paper is not indexed in the Web of Science Core Collection, how can its academic attention be tracked?

• Answer: Use the Cited Reference Search function in Web of Science by entering the author and year, rather than relying solely on "Basic Search."
• Analysis:
  This is a common retrieval misconception. Even if the article itself is not indexed by SCI/SSCI (e.g., published in unindexed conferences or monographs), as long as it has been cited by other high-level papers indexed in the database, its citation record can still be found through the "Cited Reference Search" function. This is an important means of assessing the impact of papers in non-core journals.

Q4: What are the main differences between Web of Science (WoS) and Scopus in citation tracking?

• Answer: WoS focuses on historical depth and selection criteria (earlier backfiles), while Scopus has broader coverage and includes more regional literature.
• Analysis:
  According to a comparative analysis by Journal Metrics, Web of Science was founded by Eugene Garfield and has stricter journal selection criteria, making it suitable for deep historical tracing (dating back to 1900); Scopus, on the other hand, indexes a larger number of journals and updates faster, making it suitable for broad interdisciplinary citation analysis. When answering such comparison questions, one should choose the appropriate database based on retrieval needs ("comprehensiveness" vs. "precision").

Document Codes and Chinese Library Classification (CLC)

This section mainly examines the candidate's mastery of document standards, classification systems, and retrieval languages. These questions usually belong to "hard knowledge," requiring accurate memorization of common code meanings and classification logic. In actual engineering and academic retrieval, using classification numbers (such as CLC or IPC) for retrieval is a core method to solve "keyword ambiguity" and "low precision" problems.

11. What are the meanings and differences of common document codes (ISBN, ISSN, DOI, IPC)?

Problem Analysis: This is the most basic matching question in information retrieval exams, aiming to test whether users can identify document types through codes.

Reference Answer:

• ISBN (International Standard Book Number): Used to identify books.
• ISSN (International Standard Serial Number): Used to identify continuous publications such as journals and magazines.
• DOI (Digital Object Identifier): A unique identifier for digital objects, commonly used to identify digital resources such as electronic journal papers and datasets, possessing permanent link characteristics.
• IPC (International Patent Classification): Used to identify patent documents.

Engineering Practice Tip: When writing crawlers or performing automated metadata extraction, DOI is the most reliable "primary key" because it does not become invalid like URLs due to website revisions.

Quick Reference Table: Correspondence between Codes and Document Types

12. What do the document type identification codes ([J], [M], [D]) in references represent?

Problem Analysis: Such questions usually appear in the "citation format specification" or "retrieval result interpretation" sections.

Reference Answer:
According to the GB/T 7714 standard, the single-letter identifiers in the reference list represent the type of carrier cited:

• [J] (Journal): Journal article.
• [M] (Monograph): Monograph (ordinary book).
• [D] (Dissertation): Dissertation (Doctoral or Master's).
• [P] (Patent): Patent.
• [C] (Conference): Conference proceedings.

Code Example:

"Li, Y. (2023). Deep Learning for IR. Journal of Computer Science, 10(2), 45-50. [J]"

13. What are the basic structure and common categories of the "Chinese Library Classification" (CLC)?

Problem Analysis: The CLC is the underlying classification standard for domestic databases (such as CNKI, Wanfang). Exams often test the meaning of main category letters and how to narrow the retrieval scope through classification numbers.

Reference Answer:
The "Chinese Library Classification" uses a mixed "Letter + Number" system. It divides knowledge into 5 main parts and 22 large classes. In engineering technology and academic retrieval, the following categories are the most frequent:

• T Class: Industrial Technology (Technology). This is the largest category in science and engineering retrieval.
• • TP: Automation technology, computer technology.
  • TB: General industrial technology (e.g., nanomaterials TB383).
  • TU: Building science / Architecture.
• O Class: Mathematical sciences and chemistry (Basic Sciences).
• F Class: Economics.
• I Class: Literature.
• J Class: Art.

Retrieval Technique:
When keywords (such as "optimization") have completely different meanings in different fields, limiting the CLC classification number is the best way to improve precision. For example, in Dissertation Topic Query, if the research object is "mathematical optimization," it should be limited to Class O; if researching "database query optimization," it should be limited to Class TP.

14. How to interpret the hierarchical structure of CLC numbers (taking TP311 as an example)?

Problem Analysis: Examines the candidate's understanding of classification hierarchy, i.e., "the longer the category number, the more specific the subject."

Reference Answer:
CLC numbers indicate hierarchical subdivision through appended characters. Taking the commonly used computer software TP311 as an example:

1. T: Industrial Technology (Level 1).
2. TP: Automation technology, computer technology (Level 2).
3. TP3: Computing technology, computer technology (Level 3).
4. TP31: Computer software.
5. TP311: Programming, software engineering.

Common Pitfalls:

• Accuracy of Classification Numbers: When transcribing or retrieving, attention must be paid to the combination of letters and numbers. For example, CLC Classification Number points out that inadvertently writing TP2 (Automation technology equipment) as TP1 (Automation basic theory) will lead to completely deviated retrieval results.
• Subdivision and Auxiliary Subdivision: Certain classification numbers may contain symbols like "-" or "()", indicating regional subdivision or chronological subdivision (e.g., F740.4-712 might indicate the trade economy of a specific country), which belongs to advanced examination points.

15. What is the difference between Chinese Library Classification (CLC) and Universal Decimal Classification (UDC)?

Problem Analysis: This is a comparison question aimed at advanced retrieval or international document management.

Reference Answer:

• Scope of Application: CLC is a Chinese national standard, mainly used in domestic libraries and Chinese databases like CNKI; UDC (Universal Decimal Classification) is an internationally common classification method, widely used in international document retrieval.
• Identification Differences:
• • CLC Example: TP391 (Computer Application).
  • UDC Example: 004.8 (Artificial Intelligence).
• Application Scenarios: When writing dissertations or submitting to international journals, it is usually necessary to mark both CLC and UDC classification numbers to ensure the document can be accurately indexed in both domestic and international systems.

Search Strategy Optimization: Recall and Precision

This section translates theory into practice, focusing on the ability to debug when search results are "too noisy" (too many results) or "missed" (too few results). This is the watershed distinguishing beginners from experts in information retrieval: experts not only know how to construct search queries but also understand how to dynamically adjust strategies based on feedback to find the optimal balance between Recall and Precision.

26. Scenario Question: Noise Reduction Strategies When Search Results "Explode"

Question: You are searching for literature related to "Deep Learning" applications in an engineering database, and the results return over 10,000 records, which is obviously impossible to read one by one. Please list at least three mechanical (not simply changing keywords) operations to quickly narrow the scope and improve relevance.

Answer and Analysis:
Facing massive results (High Recall, Low Precision) requires adding constraints to improve precision.

1. Field Limiting: Change the default "Topic" or "Full Text" search to be limited to the Title (TI) or Keywords (KY) fields.

• • Principle: Words appearing in the title usually represent the core theme of the article better than words appearing in the full text.
  • Example: Change Deep Learning to TI="Deep Learning".

1. Use Proximity Operators: If the database supports it, use NEAR/n or W/n instead of AND.

• • Principle: AND only guarantees word co-occurrence, not logical association; proximity operators require words to be physically close, resulting in higher relevance.

1. Add Logic "AND" Constraints: Introduce a second conceptual dimension (such as an application scenario or specific technology).

• • Example: "Deep Learning" AND "Image Recognition".

1. Utilize Facets/Filters: Perform secondary filtering in the result page sidebar by "Document Type (Review/Article)", "Publication Year", or "Subject Category".

Note: South China Normal University's Search Skills Guide points out that trial searches via different search fields is one of the most direct strategies to expand or narrow the scope.

27. Scenario Question: Expansion Strategies for Zero or Few Results

Question: You are researching an emerging niche field (e.g., "Photonic crystal-based biosensors") and have found only 2 papers after a preliminary search, with low relevance. Please describe how to adjust the search query to improve Recall.

Answer and Analysis:
Too few results (Low Recall) usually means the search constraints are too rigid or the vocabulary fails to cover all expressions.

1. Synonym Expansion and Logic "OR": List synonyms, abbreviations, and variants of core concepts.

• • Operation: Expand Biosensor to (Biosensor OR "Biological sensor" OR Bio-sensor).

1. Use Truncation: Use * or ? to cover suffix variations.

• • Operation: Using Photonic crystal* can simultaneously retrieve forms like crystal, crystals, and crystalline.

1. Reduce Field Limits: If TI= was previously limited, relax it to AB= (Abstract) or SU= (Subject), or even full-text search.
2. Broader Term Search: If the specific concept is too new, try searching for its Broader Terms, then manually screen the results.

28. Conceptual Question: The Trade-off Between Recall vs. Precision

Question: In the two scenarios of "Patent Infringement Search" and "Daily Technical Q&A", how does your pursuit of recall and precision differ? Please explain briefly combined with definitions.

Answer and Analysis:

• Recall = Number of relevant documents retrieved / Total number of relevant documents in the database.
• Precision = Number of relevant documents retrieved / Total number of documents retrieved.

The two usually show a negative correlation:

• Patent Infringement/Medical Systematic Review: Recall is paramount. Missing a key piece of prior art patent could lead to huge legal damages or wasted R&D. In this case, one should tolerate a lot of noise (low precision) and use complex Boolean logic OR to stack synonyms.
• Daily Technical Q&A/Quick Verification: Precision comes first. Engineers just want to quickly find a usable solution (e.g., "Python array deduplication method") and do not need to read all relevant historical literature. In this case, exact matching and field limiting should be used to ensure the first few results are what is needed.

29. Advanced Question: Controlled Vocabulary vs. Natural Language

Question: When using professional databases like Ei Compendex or PubMed, why does inputting precise English words sometimes result in missing important literature? Please explain the role of "Controlled Terms" and their difference from natural language search.

Answer and Analysis:
This is because Natural Language has polysemy and synonymy, while professional databases normalize concepts through a Thesaurus.

• Phenomenon: You search for Cell phone and might miss early core literature that uses Mobile phone or Cellular telephone.
• Controlled Terms (Subject Headings): Database indexers will tag all literature representing "mobile phones" with a standard label (e.g., Mobile telecommunication systems).
• Strategy:

1. 1. Search with natural language keywords first.
   2. Check the Controlled Terms / Subject Headings field of highly relevant hits.
   3. Click these controlled terms to perform a "secondary search", which usually significantly improves recall and excludes irrelevant homonyms.

According to the Ei Compendex Database Guide, utilizing the Thesaurus to find standard terms in a specific research field can effectively avoid missed checks due to different spellings or false checks caused by conceptual confusion.

30. Modern Perspective: Search "Hallucinations" and Verification in the AI Era

Question: When using ChatGPT or similar generative AI tools for technical retrieval (e.g., "List review papers on Transformer architecture from 2023"), what is the most common risk? How should an engineer verify this?

Answer and Analysis:

• Risk: Hallucination. Generative models predict the next character based on probability rather than querying a real-time index. They may fabricate non-existent paper titles, authors, or even DOIs that look very realistic but are completely fictional.
• Verification Strategy:

1. 1. Cross-validation: Put the AI-generated paper titles into Google Scholar or IEEE Xplore for review to confirm the literature actually exists.
   2. Request Links: Ask the AI to provide specific URLs (although URLs can also be fabricated, they are easier to verify).
   3. Use Connected/RAG Tools: Use AI tools integrated with search engines (such as Bing Chat or Perplexity), which are based on Retrieval-Augmented Generation (RAG) technology and will cite actual crawled web snippets when generating content, significantly reducing the risk of hallucinations.