Abstract:E-recruitment recommendation systems recommend jobs to job seekers and job seekers to recruiters. The recommendations are generated based on the suitability of job seekers for positions and on job seekers’ and recruiters’ preferences. Therefore, e-recruitment recommendation systems may greatly impact people’s careers. Moreover, by affecting the hiring processes of the companies, e-recruitment recommendation systems play an important role in shaping the competitive edge of companies. Hence, it seems prudent to consider what (unique) challenges there are for recommendation systems in e-recruitment. Existing surveys on this topic discuss past studies from the algorithmic perspective, e.g., by categorizing them into collaborative filtering, content-based, and hybrid methods. This survey, instead, takes a complementary, challenge-based approach. We believe this is more practical for developers facing a concrete e-recruitment design task with a specific set of challenges, and also for researchers that look for impactful research projects in this domain. In this survey, we first identify the main challenges in the e-recruitment recommendation research. Next, we discuss how those challenges have been studied in the literature. Finally, we provide future research directions that we consider most promising in the e-recruitment recommendation domain.
What problem does this paper attempt to address?
### Problems the Paper Attempts to Solve
The paper attempts to address the unique challenges faced by e-recruitment recommendation systems. Specifically, the authors review the literature on e-recruitment recommendation systems over the past 10 years using a challenge-based approach. Unlike previous reviews that primarily classify research methods from an algorithmic perspective (such as collaborative filtering, content-based recommendation, and hybrid methods), this paper focuses on specific challenges in e-recruitment recommendation systems and their solutions.
### Main Challenges
1. **Data Quality**:
- **Data Cleaning and Preprocessing**: Recommendation systems typically use features extracted from textual data (such as resumes and job descriptions), which are often written by different people and are noisy, making data cleaning and preprocessing crucial.
- **Semantic Gap**: Due to textual data being written by different people, e-recruitment recommendation systems face a semantic gap between content from different sources (such as resumes and job descriptions).
- **Skill Extraction**: Extracting skills from textual data is an important task for e-recruitment recommendation systems, as skill matching is crucial for improving recommendation performance.
- **Multilingual Support**: In some countries or platforms, resumes and job descriptions may be written in multiple languages, requiring recommendation systems to support multilingual textual content.
- **Data Sparsity**: Many recommendation systems face the problem of data sparsity, and e-recruitment recommendation systems are no exception, as job seekers may leave the platform after only a few uses, and new job positions may frequently appear and disappear.
2. **Heterogeneous Data and Multiple Interaction Types and Data Sources**:
- E-recruitment recommendation systems can use more data sources, including job seekers' past work experience, interview records, textual content of resumes and job descriptions, skills, and preferences.
- There are multiple interaction types in the system, such as viewing, clicking, applying, chatting, bookmarking, liking, and commenting, which bring both challenges and opportunities.
3. **Cold Start Problem**:
- The cold start problem is common in recommendation systems, where there is a lack of sufficient interaction data for new users or new items. This problem is particularly prominent in e-recruitment recommendation systems, as new job positions frequently appear and disappear.
4. **User Preferences and Adaptability**:
- When finding the best match for job seekers and positions, it is necessary to consider not only the job seekers' knowledge and skills but also their personal preferences.
5. **Explainability and Transparency**:
- Providing explainable recommendations and designing explainable models are very important in e-recruitment recommendation systems, as important career decisions rely on these recommendations.
6. **Specific Objectives**:
- E-recruitment recommendation systems often have multi-objective nature, needing to meet the needs of multiple stakeholders, such as job seekers, recruiters, and service providers.
- The system may also have specific objectives, such as balancing the number of recommendations each job seeker or position receives or recommending items with a higher likelihood of success.
7. **Bias and Fairness**:
- Recommendation systems have various known biases, some of which raise social and ethical concerns. In the high-stakes field of e-recruitment, providing fair recommendations is particularly important, requiring mitigation of issues such as gender bias and recency bias.
8. **Scalability**:
- With the continuous increase in data volume, scalability becomes an urgent challenge for e-recruitment recommendation systems. Large-scale data may bring speed and storage/memory consumption issues during training and inference stages.
### Contributions
- Emphasized the importance of specialized research on e-recruitment recommendation systems and listed some important characteristics of this field.
- Identified and briefly discussed eight common challenges and their solutions.
- For each challenge, discussed research papers specifically addressing that challenge and their methods.
- Provided future research directions and discussed challenges that have been less studied in recent years.
- Provided a structured literature overview table (Table 1 in the appendix), listing the recommendation type, recommendation method type, and challenges addressed by each paper.
- Maintained a website containing the contents of Table 1, as well as metadata of the papers (such as publication venue, URL, authors, etc.) and abstracts of selected papers.
### Conclusion
The paper...