The within-basket recommendation task refers to recommending items to complete the current incomplete basket. Previous nearest neighbor-based methods implicitly assume that the set of neighbors remains constant across all items. However, user preferences can fluctuate based on item categories or specific occasions. In this paper, we propose PINN, a neighborhood-based model, which searches distinct sets of neighbors for different items by considering personalized co-occurrence frequency information which balances short-term and long-term preferences for within-basket tasks. PINN incorporates five key factors: item-level user similarity, item loyalty, item popularity, recency, and co-occurrence items. Extensive experiments were conducted on six public datasets of varying scales. Results demonstrate that PINN consistently outperforms state-of-the-art baselines across different datasets. We share our code at https://github.com/Daisy-24/PINN .